A method for smooth switching cooperative adaptive cruise control with dynamic topology

Abstract

The application discloses a smooth switching cooperative adaptive cruise control method with a dynamic topology, and comprises the following steps: designing a dynamic topology following information flow architecture to provide scene adaptive information interaction support for a cooperative adaptive cruise control system of an automatic driving vehicle queue; constructing a topology smooth switcher based on fuzzy logic to realize stepless smooth switching of different information flow topologies through dynamic adjustment of the weight of multiple information sources; and adopting a phased progressive training strategy to complete construction and optimization of the queue cooperative cruise controller based on a multi-agent proximal policy optimization algorithm; the application can effectively improve stability, robustness and safety of automatic driving vehicle queue driving, improve road traffic efficiency and vehicle driving economy, and has great significance for engineering landing of automatic driving vehicle queue technology and large-scale application of an intelligent transportation system.

Description

Technical Field

[0001] This invention belongs to the field of autonomous vehicle platoon control technology, specifically relating to a smooth switching cooperative adaptive cruise control method with dynamic topology, used to solve the problem of unstable control caused by information flow topology switching in non-ideal communication environments. Background Technology

[0002] With the rapid development of autonomous driving technology, autonomous vehicle platoons (also known as "intelligent connected vehicle fleets") have become one of the key technologies for improving road traffic efficiency, alleviating traffic congestion, and reducing energy consumption and accident risks. Cooperative adaptive cruise control, as the core technology for achieving cooperative operation of vehicle platoons, obtains information from multiple vehicles ahead through vehicle-to-vehicle (V2V) wireless communication, enabling the platoon to achieve stable and efficient formation driving with a smaller safety distance. This allows more vehicles to be accommodated in the same road space, improving traffic efficiency.

[0003] However, most existing cooperative adaptive cruise control systems are based on fixed information flow topologies (such as leading vehicle following topology, guide vehicle bidirectional following topology, etc.). This fixed topology design has significant limitations: First, it cannot adaptively select the optimal information interaction mode based on real-time vehicle driving conditions (such as vehicle spacing error) and dynamically changing communication environments (such as communication packet loss rate). When facing complex and ever-changing traffic conditions, fixed topologies either suffer from insufficient information leading to decreased control performance, or increase the communication burden and reduce the system's robustness to single-point communication failures due to information redundancy.

[0004] Secondly, to address different scenarios, some existing technologies attempt to switch between different fixed topologies. However, this switching is usually a "hard switch," meaning an instantaneous jump from one topology to another. Hard switches cause sudden changes in the information source, leading to violent fluctuations in control commands, resulting in acceleration oscillations, which seriously affect the smoothness of vehicle platooning and passenger comfort, and may even cause safety hazards.

[0005] Furthermore, in the training of cooperative adaptive cruise controllers, traditional single-stage training methods attempt to simultaneously optimize multiple objectives such as safety, efficiency, and comfort. Due to the inherent coupling and conflicts between these objectives, this parallel optimization approach is prone to causing the training process to fail to converge, or the ultimately learned control strategy to exhibit mediocre performance across all aspects, failing to meet the high standards of safety, efficiency, and comfort required by platoons of autonomous vehicles.

[0006] Therefore, there is an urgent need in this field for a cooperative adaptive cruise control method that can overcome the above-mentioned defects and enable vehicle platoons to operate safely, smoothly and efficiently even in non-ideal communication environments. Summary of the Invention

[0007] To overcome the shortcomings of the prior art, the purpose of this invention is to provide a smooth switching cooperative adaptive cruise control method with dynamic topology. Based on a multi-agent near-end policy optimization algorithm, it solves the problems of poor adaptability of fixed information flow topology in dynamic scenarios, non-smooth control caused by hard topology switching, and difficulty in training multi-objective controllers. It can effectively improve the longitudinal and lateral cooperative control performance of automatic vehicle platoons in non-ideal communication environments.

[0008] To achieve the above objectives, the technical solution adopted by this invention is: a smooth switching cooperative adaptive cruise control method with dynamic topology, comprising the following steps:

[0009] Step S1: Design a dynamic topology following information flow architecture to provide scene-adaptive information interaction support for the autonomous vehicle platoon cooperative adaptive cruise control system; the dynamic topology following information flow architecture includes three topology modes: following the preceding vehicle, following the leading vehicle, and following the leading vehicle in both directions, and uses platoon spacing error and communication packet loss rate as two-dimensional decision criteria to achieve adaptive dynamic matching of the topology.

[0010] Step S2: Construct a topology smoothing switcher based on fuzzy logic to achieve seamless and smooth switching of different information flow topologies through dynamic adjustment of multi-source information weights; the input variables of the topology smoothing switcher are spacing error and communication packet loss rate, and the output variables are the information weight adjustment amounts of the preceding vehicle, the guiding vehicle, and the following vehicle.

[0011] Step S3: A phased progressive training strategy is adopted to complete the construction and optimization of the queue cooperative cruise controller based on the multi-agent proximal policy optimization algorithm; the phased progressive training strategy includes at least a distance tracking stage, a velocity tracking stage, and an acceleration tracking stage.

[0012] 2. The method for smooth switching cooperative adaptive cruise control with dynamic topology according to claim 1, characterized in that, in step S1, the communication packet loss rate is divided into three levels: 5% for low packet loss rate, 10% for medium packet loss rate, and 20% for high packet loss rate.

[0013] 3. The method for smooth switching cooperative adaptive cruise control with dynamic topology according to claim 1, characterized in that, in step S2, constructing a topology smooth switcher based on fuzzy logic specifically includes: dividing the spacing error into three fuzzy subsets (large, medium, and small), and dividing the communication packet loss rate into three fuzzy subsets (low, medium, and high); using a triangular membership function to map precise input values ​​to the fuzzy subsets; constructing a fuzzy rule base containing 9 core rules; and using the centroid method for defuzzification to convert the fuzzy output set into precise weight adjustment amounts.

[0014] 4. The method for smooth switching cooperative adaptive cruise control with dynamic topology according to claim 3, characterized in that the nine core rules are:

[0015] Rule 1: If the spacing error is medium and the communication packet loss rate is low, then the weight of the preceding vehicle is low, the weight of the guide vehicle is high, and the weight of the following vehicle is low.

[0016] Rule 2: If the spacing error is medium and the communication packet loss rate is medium, then the weight of the preceding vehicle is low, the weight of the guide vehicle is high, and the weight of the following vehicle is low.

[0017] Rule 3: If the spacing error is medium and the communication packet loss rate is high, then the weight of the preceding vehicle is medium, the weight of the guide vehicle is medium, and the weight of the following vehicle is low.

[0018] Rule 4: If the spacing error is large and the communication packet loss rate is low, then the weight of the preceding vehicle is low, the weight of the guide vehicle is low, and the weight of the following vehicle is high.

[0019] Rule 5: If the spacing error is large and the communication packet loss rate is medium, then the weight of the preceding vehicle is low, the weight of the guide vehicle is medium, and the weight of the following vehicle is medium.

[0020] Rule 6: If the spacing error is large and the communication packet loss rate is high, then the weight of the preceding vehicle is low, the weight of the guide vehicle is high, and the weight of the following vehicle is low.

[0021] Rule 7: If the spacing error is small and the communication packet loss rate is low, then the weight of the preceding vehicle is high, the weight of the guide vehicle is low, and the weight of the following vehicle is low.

[0022] Rule 8: If the spacing error is small and the communication packet loss rate is medium, then the weight of the preceding vehicle is high, the weight of the guide vehicle is low, and the weight of the following vehicle is low.

[0023] Rule 9: If the spacing error is small and the communication packet loss rate is high, then the weight of the preceding vehicle is high, the weight of the guide vehicle is low, and the weight of the following vehicle is low.

[0024] 5. The method for smooth switching cooperative adaptive cruise control with dynamic topology according to claim 1, characterized in that the phased progressive training strategy in step S3 specifically comprises:

[0025] Distance tracking phase: A fixed leading vehicle following topology is adopted, and the smooth switching function is disabled. The training objective focuses on maintaining a safe distance. The total reward function for this phase is based on formula (1). for:

[0026] (1)

[0027] in, For vehicle spacing tracking reward items, This is a collision hard constraint penalty term. This refers to the weighting coefficient for the vehicle spacing reward item;

[0028] Speed ​​tracking phase: The topology smooth switching function is enabled, but switching is only allowed between the leading vehicle following and the guide vehicle following topologies. The training objective is to increase speed synchronization while maintaining a safe distance. The total reward function for this phase is formula (4):

[0029] (4)

[0030] in, The weighting coefficient for the speed reward item. This is a newly added speed tracking reward item;

[0031] Acceleration tracking phase: Allows switching between the entire topology, and the training objective further increases the acceleration smoothness. The total reward function for this phase is Equation (6):

[0032] (6)

[0033] in, The weighting coefficient for acceleration reward items. This is a newly added acceleration smoothing bonus item.

[0034] 6. The method for smooth switching cooperative adaptive cruise control with dynamic topology according to claim 5, characterized in that the vehicle spacing tracking reward term in the total reward function of the distance tracking phase adopts formula (2):

[0035] (2)

[0036] in, To account for the following vehicle's spacing error, This is the spacing error gain coefficient;

[0037] The collision hard constraint penalty term adopts formula (3):

[0038] (3)

[0039] in, This is the maximum penalty value.

[0040] 7. A smooth switching cooperative adaptive cruise control method with dynamic topology according to claim 5, characterized in that the speed tracking reward term in the total reward function of the speed tracking phase adopts formula (5):

[0041] (5)

[0042] in, For speed error, This is the speed error gain coefficient.

[0043] 8. A smooth switching cooperative adaptive cruise control method with dynamic topology according to claim 5, characterized in that the acceleration smoothing reward term in the total reward function of the acceleration tracking phase adopts formula (7):

[0044] (7)

[0045] in, To provide real-time acceleration for following the vehicle, The gain coefficient corresponding to the acceleration amplitude. The gain coefficient corresponding to the rate of change of acceleration. This refers to the rate of change of acceleration in real time following the vehicle.

[0046] Compared with the prior art, the cooperative adaptive cruise control method and system with dynamic topology smooth switching function provided by the present invention has the following significant advantages:

[0047] 1) Significantly improved robustness and adaptability of vehicle platooning: By dynamically adjusting the information flow topology based on real-time vehicle spacing error and communication packet loss rate (a dynamic topology-following information flow design was implemented), this invention frees vehicle platooning from a single fixed communication mode, providing scenario-adaptive information interaction support for the cooperative adaptive cruise control system. The system can sense and adapt to communication quality fluctuations and vehicle state changes, automatically switching to a better mode when the topology structure does not match the scenario requirements. This ensures that the platooning maintains stability and control performance in various complex and non-ideal driving environments (such as urban roads, tunnels, and severe weather), effectively avoiding the risk of platooning instability caused by communication interruptions or sudden state changes, thus significantly enhancing robustness.

[0048] 2) Fundamentally eliminates the control shock caused by topology "hard switching," greatly improving driving smoothness and ride comfort: The innovative smooth switching mechanism of this invention achieves seamless transition of control by dynamically fusing control commands from the old and new topologies through a fuzzy logic controller. This method effectively suppresses the acceleration abrupt changes and vehicle "nodding" phenomenon caused by the instantaneous switching of information sources in traditional hard switching, making the acceleration and deceleration processes of each following vehicle in the platoon smoother and more fluid, significantly improving driving comfort and providing a guarantee for the long-term safe operation of the platoon. This invention achieves stepless smooth switching of the topology through dynamic adjustment of multi-source information weights, effectively suppressing acceleration oscillations during the switching process and improving the ride smoothness of the vehicle platoon.

[0049] 3) By employing a phased reinforcement learning strategy, the bottleneck of multi-objective controller optimization is overcome, achieving global optimality in control performance. The complex multi-objective optimization problems, such as maintaining safe distances, speed tracking, and comfort, are decomposed into hierarchical, progressive training phases, reducing the learning difficulty for the agent and the complexity of the policy search space. This method not only accelerates the convergence speed of reinforcement learning but, more importantly, guides the agent to prioritize learning core skills that ensure safety (such as collision avoidance) before gradually optimizing higher-order performance (such as comfort). The resulting control policy achieves an excellent and balanced Pareto optimal state across all key performance indicators (efficiency and comfort as optimization objectives). The decoupling design avoids training oscillations caused by multi-objective coupling, accelerating policy convergence and improving learning efficiency. It overcomes the shortcomings of traditional single-stage training, which is prone to getting trapped in local optima or performance compromises.

[0050] 4) Optimized communication resource utilization and improved system efficiency: The dynamic topology management mechanism can intelligently select the most concise and effective information interaction mode while ensuring control performance, avoiding unnecessary communication load, thereby saving energy consumption and channel resources of the vehicle communication unit to a certain extent and improving the operating efficiency of the entire queue system. Attached Figure Description

[0051] Figure 1 This is a flowchart of the method of the present invention.

[0052] Figure 2 This is a schematic diagram of the dynamic topology following information flow structure of the present invention.

[0053] Figure 3 This is a framework diagram of the fuzzy logic topology smoothing switcher of the present invention.

[0054] Figure 4 This is a framework diagram of the multi-agent proximal policy optimization algorithm of the present invention.

[0055] Figure 5 This is a diagram of the phased progressive training framework of the present invention.

[0056] Figure 6 The graphs show the acceleration and velocity curves of the guide vehicle under three working conditions according to the present invention.

[0057] Figure 7 This is a graph showing the average reward curve during the training process of the three information flow topologies of this invention.

[0058] Figure 8 This is a graph showing the acceleration and velocity of four following vehicles under three working conditions in the dynamic topology following method of this invention.

[0059] Figure 9This is a comparison chart of the queuing speed error of autonomous vehicles under three different information flow topologies in the present invention.

[0060] Figure 10 This is a comparison chart of the queue spacing error of autonomous vehicles under three different information flow topologies in the three working conditions of this invention.

[0061] Figure 11 This is an error diagram showing the speed difference of autonomous vehicle queues relative to ideal communication under three working conditions of the present invention, with different communication packet loss rates and information flow topologies.

[0062] Figure 12 This is an error diagram showing the difference in queue spacing between autonomous vehicles relative to ideal communication under three different communication packet loss rates and information flow topologies in the three working conditions of this invention.

[0063] Figure 13 This is a comparison chart of the maximum collision time and the deceleration rate for avoiding collisions for three information flow topologies under different communication packet loss rates according to the present invention.

[0064] Figure 14 This is a flowchart of the hardware-in-the-loop experimental closed-loop control of the present invention.

[0065] Figure 15 This is a comparison chart of simulation and experimental results under working condition 1 of the present invention.

[0066] Figure 16 This is a comparison chart of simulation and experimental results under working condition 2 of the present invention.

[0067] Figure 17 This is a comparison chart of simulation and experimental results under working condition 3 of the present invention. Detailed Implementation

[0068] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. The embodiments described below are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] Example 1

[0070] See Figure 1-2 A smooth-switching cooperative adaptive cruise control method with dynamic topology, characterized by comprising the following steps:

[0071] Step S1: Design a dynamic topology following information flow architecture to provide scene-adaptive information interaction support for the autonomous vehicle platoon cooperative adaptive cruise control system; the dynamic topology following information flow architecture includes three topology modes: following the preceding vehicle, following the leading vehicle, and following the leading vehicle in both directions, and uses platoon spacing error and communication packet loss rate as two-dimensional decision criteria to achieve adaptive dynamic matching of the topology.

[0072] Step S2: Construct a topology smoothing switcher based on fuzzy logic to achieve seamless and smooth switching of different information flow topologies through dynamic adjustment of multi-source information weights; the input variables of the topology smoothing switcher are spacing error and communication packet loss rate, and the output variables are the information weight adjustment amounts of the preceding vehicle, the guiding vehicle, and the following vehicle.

[0073] Step S3: A phased progressive training strategy is adopted to complete the construction and optimization of the queue cooperative cruise controller based on the multi-agent proximal policy optimization algorithm; the phased progressive training strategy includes at least a distance tracking stage, a velocity tracking stage, and an acceleration tracking stage.

[0074] Step 1, which involves designing a dynamic topology to follow the information flow, is specifically established as follows:

[0075] In cooperative adaptive cruise control scenarios, traditional autonomous vehicle platoons typically employ a fixed information flow topology. However, the bidirectional following topology of the lead vehicle is prone to increased communication load due to information redundancy, and the leading vehicle following topology has weak fault tolerance for single-node failures, making it difficult to meet the cooperative requirements of autonomous vehicle platoon scenarios. To address these issues, this study proposes a dynamic topology following topology, see [link to relevant documentation]. Figure 2 As shown, its core is to dynamically match the topology with the queue status and communication environment of autonomous vehicles by adaptively adjusting the weight of information sources.

[0076] Based on the information exchange requirements of autonomous vehicle platoons, three topology modes were designed: leading vehicle following, guide vehicle following the leading vehicle, and guide vehicle following in both directions. Each mode corresponds to a different set of information sources, achieving lightweight adaptation of information exchange. Topology switching follows the core logic of safety first and communication adaptiveness, using platoon spacing error and communication packet loss rate as two-dimensional decision criteria to avoid blind topology adjustments.

[0077] In step S1, based on the actual vehicle communication scenario, the communication packet loss rate is divided into three levels: 5% is a low packet loss rate, 10% is a medium packet loss rate, and 20% is a high packet loss rate.

[0078] Compared with traditional fixed communication topologies, the proposed dynamic topology following topology can autonomously select the optimal information interaction mode according to the dynamic changes in driving conditions and communication environment, breaking the limitation of fixed topologies in adapting to a single working condition.

[0079] In step S2, a topology smoothing switcher based on fuzzy logic is constructed, specifically as follows:

[0080] In cooperative adaptive cruise control systems, dynamic topology enhances the adaptability of information interaction. However, traditional hard topology switching often causes abrupt information changes, leading to fluctuations in the following vehicle control commands and consequently acceleration oscillations, significantly reducing cruise stability. Therefore, this invention proposes a fuzzy logic topology smoothing switcher. Through input fuzzification, construction of an empirical rule base, and dynamic adjustment of information source weights, it achieves a stepless smooth transition in topology switching, effectively compensating for the rigidity defects of traditional hard switching.

[0081] This switcher transforms the discrete switching between leading vehicle following, guide vehicle following, and guide vehicle bidirectional following into a continuous dynamic adjustment of information source weights, avoiding abnormal fluctuations in control commands caused by sudden information changes. Simultaneously, it fuzzifies queue status and communication packet loss rate, solving the problem that such information cannot be precisely quantified using fixed thresholds.

[0082] The input layer uses spacing error and packet loss rate as input variables to reflect the reliability of the communication link. In the fuzzification step, the precise input is converted into fuzzy variables, the spacing error is divided into three fuzzy subsets (large, medium, and small), and the packet loss rate is divided into three fuzzy subsets (low, medium, and high). A triangular membership function is used to map the precise input values ​​to the fuzzy subsets, ensuring the rationality and continuity of the fuzzification.

[0083] The output layer focuses on adjusting the weights of information sources, achieving a smooth transition in the topology through continuous weight allocation. Output variables include the weight adjustments for the preceding vehicle, the lead vehicle, and the following vehicles. For example, when switching from a topology where the preceding vehicle follows the lead vehicle to one where the preceding vehicle follows the lead vehicle, the weight adjustment for the lead vehicle is positive, gradually increasing the proportion of the lead vehicle's information, ultimately achieving a smooth and continuous transition in the topology.

[0084] The rule base is the core component of the smooth switcher. Based on engineering experience and safety requirements of cooperative adaptive cruise control, a fuzzy rule base containing nine core rules is constructed, covering all combinations of the input fuzzy subset to ensure that reasonable weight adjustment instructions are given under different scenarios. The centroid method is used for defuzzification, converting the fuzzy output set into precise weight adjustment values.

[0085] The nine core rules are as follows:

[0086] Rule 1: If the spacing error is medium and the communication packet loss rate is low, then the weight of the preceding vehicle is low, the weight of the guide vehicle is high, and the weight of the following vehicle is low.

[0087] Rule 2: If the spacing error is medium and the communication packet loss rate is medium, then the weight of the preceding vehicle is low, the weight of the guide vehicle is high, and the weight of the following vehicle is low.

[0088] Rule 3: If the spacing error is medium and the communication packet loss rate is high, then the weight of the preceding vehicle is medium, the weight of the guide vehicle is medium, and the weight of the following vehicle is low.

[0089] Rule 4: If the spacing error is large and the communication packet loss rate is low, then the weight of the preceding vehicle is low, the weight of the guide vehicle is low, and the weight of the following vehicle is high.

[0090] Rule 5: If the spacing error is large and the communication packet loss rate is medium, then the weight of the preceding vehicle is low, the weight of the guide vehicle is medium, and the weight of the following vehicle is medium.

[0091] Rule 6: If the spacing error is large and the communication packet loss rate is high, then the weight of the preceding vehicle is low, the weight of the guide vehicle is high, and the weight of the following vehicle is low.

[0092] Rule 7: If the spacing error is small and the communication packet loss rate is low, then the weight of the preceding vehicle is high, the weight of the guide vehicle is low, and the weight of the following vehicle is low.

[0093] Rule 8: If the spacing error is small and the communication packet loss rate is medium, then the weight of the preceding vehicle is high, the weight of the guide vehicle is low, and the weight of the following vehicle is low.

[0094] Rule 9: If the spacing error is small and the communication packet loss rate is high, then the weight of the preceding vehicle is high, the weight of the guide vehicle is low, and the weight of the following vehicle is low.

[0095] Finally, the centroid method is used for defuzzification, converting the fuzzy output set obtained from fuzzy inference into precise weight adjustment values. By calculating the centroid coordinates of the fuzzy set as the output value, a smooth adjustment of the information source weights is achieved without abrupt changes.

[0096] The phased progressive training strategy in step S3 is specifically as follows:

[0097] In cooperative adaptive cruise control training, traditional single-stage training methods are prone to coupling conflicts when optimizing multiple objectives in parallel. Therefore, this study proposes a phased, progressive training strategy. Training objectives are decomposed according to the priorities of safety, efficiency improvement, and comfort optimization, while gradually increasing topological complexity and environmental disturbances, enabling the agent to progressively learn cooperative cruise skills. (See [link to relevant documentation]). Figure 3 As shown.

[0098] The training objectives of cooperative adaptive cruise control have inherent priorities: safe distance is a basic prerequisite, speed synchronization is the core of cruise efficiency, and smooth acceleration determines ride comfort. Each training phase focuses on only a single objective, and the priority of each objective is strengthened by adjusting the weights of the reward function. The next objective is added only after the current phase objective converges, ensuring that the agent can gradually master cooperative control skills in a progressive manner.

[0099] Based on the characteristics of autonomous vehicle platooning scenarios, the training is divided into three progressive stages: distance tracking, velocity tracking, and acceleration tracking. The core configurations, training priorities, and module adaptations for each stage are as follows:

[0100] Phase 1 is the distance tracking phase: Prioritize mastering the skill of maintaining a safe following distance, avoiding collision risks, weakening speed and acceleration constraints, adopting a fixed leading vehicle following topology, disabling smooth switching functions, and focusing the training objective on maintaining a safe distance. The total reward function for this phase uses formula (1). for:

[0101] (1)

[0102] in, For vehicle spacing tracking reward items, This is a collision hard constraint penalty term. This refers to the weighting coefficient for the vehicle spacing reward item;

[0103] Phase 2 is the speed tracking phase: Based on a stable safe distance, speed synchronization with the lead vehicle is achieved to improve cruising efficiency. The topology smooth switching function is enabled, but switching is only allowed between the lead vehicle following topology and the lead vehicle following the lead vehicle topology. The training objective is to increase speed synchronization while maintaining a safe distance. The total reward function for this phase is formula (4):

[0104] (4)

[0105] in, The weighting coefficient for the speed reward item. This is a newly added speed tracking reward item;

[0106] Phase 3 is the acceleration tracking phase: controlling acceleration fluctuations to achieve smooth following, adapting to complex communication interference, completing the three-objective collaborative optimization, allowing switching between the entire topology, and further increasing acceleration smoothness in the training objective. The total reward function for this phase adopts formula (6):

[0107] (6)

[0108] in, The weighting coefficient for acceleration reward items. This is a newly added acceleration smoothing bonus item.

[0109] The vehicle spacing tracking reward term in the total reward function of the distance tracking stage adopts formula (2):

[0110] (2)

[0111] in, To account for the following vehicle's spacing error, This is the spacing error gain coefficient;

[0112] The collision hard constraint penalty term adopts formula (3):

[0113] (3)

[0114] in, This is the maximum penalty value.

[0115] The speed tracking reward term in the total reward function of the speed tracking phase adopts formula (5):

[0116] (5)

[0117] in, For speed error, This is the speed error gain coefficient.

[0118] The acceleration smoothing reward term in the total reward function of the acceleration tracking phase adopts formula (7):

[0119] (7)

[0120] in, To provide real-time acceleration for following the vehicle, The gain coefficient corresponding to the acceleration amplitude. The gain coefficient corresponding to the rate of change of acceleration. This refers to the rate of change of acceleration in real time following the vehicle.

[0121] To verify the effectiveness of the algorithm constructed in this invention, a model was built using MATLAB / Simulink and code was written. The performance of the controller was studied and analyzed under three typical disturbance conditions: ramp, sine, and step. See [link to relevant documentation]. Figure 6 As shown; and a multi-agent training framework for autonomous vehicle platooning is constructed, see [link to relevant documentation]. Figure 5 As shown, an unmanned vehicle experimental platform based on a robot operation platform was built. Under the same network segment connected by WI-FI, the longitudinal acceleration output by the multi-agent near-end strategy optimization controller was transmitted to the unmanned vehicle in the form of a message via a Topic to realize the update of the vehicle's motion state. Finally, the effectiveness of the proposed controller was verified in a real environment.

[0122] To verify the effectiveness of the constructed controller, the present invention validated the designed controller in both a simulation environment and an actual experimental platform. The results show that the control method performs well under different operating conditions. Figure 7The figures show the average reward curves during the training process of the three information flow topologies. It can be seen that the dynamic topology following topology designed in this invention, which employs phased progressive training, significantly outperforms the bidirectional following topology of the lead vehicle and the preceding vehicle following topology in terms of convergence speed and final convergence reward.

[0123] Figure 8 The figures show the acceleration and velocity of the four following vehicles under three operating conditions in the dynamic topology following topology. It can be seen that the cooperative adaptive cruise control strategy proposed under the dynamic topology following topology enables the four following vehicles to achieve accurate and stable tracking of the lead vehicle's driving state. The speeds stabilize within 10 seconds, and after stabilization, the speed difference is less than 0.2 m / s, with the speed difference almost zero in the final stage. Under all three operating conditions, the speed curves of each following vehicle quickly match the changing trend of the lead vehicle, with no significant steady-state tracking error.

[0124] Figure 9 and Figure 10 The speed and spacing errors of different information flow topologies under three different operating conditions are compared. It can be seen that the peak speed error generated by the dynamic topology designed in this patent converges rapidly within 10 seconds without significant secondary fluctuations, and the peak spacing error is less than 5m. Under operating condition 3, the dynamic topology converges more rapidly during acceleration and constant speed phases, and the error curves of multiple vehicles highly overlap, ensuring better multi-vehicle coordination consistency.

[0125] Figure 11 and Figure 12 The results show robustness analysis under non-ideal communication environments. It can be seen that as the packet loss rate increases from 5% to 20%, the performance of all three topologies decreases to varying degrees. Nevertheless, the dynamic topology maintains superior tracking accuracy and multi-vehicle coordination consistency across all operating conditions and packet loss rates. Even under the extreme condition of a 20% packet loss rate, the peak error of the dynamic topology is controlled within 0.6 m / s, 0.4 m / s, and 0.6 m / s, respectively, and the peak error of the spacing deviation is within 1.6 m, 1.6 m, and 1.2 m, with no significant performance degradation.

[0126] Figure 13This paper compares the maximum collision time and collision avoidance deceleration for three information flow topologies under different communication packet loss rates. Maximum collision time statistics show that dynamic topology following achieves the highest value in all scenarios, significantly outperforming both leading vehicle following and bidirectional guiding vehicle following. Under ideal communication conditions, the maximum collision time values ​​for dynamic topology following in the three operating conditions are 80.7, 74.4, and 78.2 seconds, respectively. As the communication packet loss rate increases from 5% to 20%, the maximum collision time for both leading vehicle following and bidirectional guiding vehicle following decreases significantly, dropping to 64.6 seconds in operating condition 2 with a 20% communication packet loss rate. In contrast, the maximum collision time for dynamic topology following remains above 75 seconds in all cases, even reaching 79.4 seconds in operating condition 3 with a 20% communication packet loss rate, demonstrating sufficient collision safety margin even in harsh communication environments.

[0127] Collision avoidance deceleration statistics show that dynamic topology following maintains a minimum value in all scenarios. Under ideal communication conditions, the collision avoidance deceleration value of dynamic topology following is 3.9 × 10⁻⁶ under all three operating conditions. -3 The deceleration rate is significantly lower than that of the vehicle following the lead vehicle and the vehicle following the lead vehicle in both directions, reaching 14.7 × 10 m / s², which is significantly lower than that of the vehicle following the lead vehicle and the vehicle following the lead vehicle in both directions, with the communication packet loss rate increasing. -3 m / s² and 16.4×10 -3 m / s², while the collision avoidance deceleration of dynamic topology following remains stable at 3.9 × 10 m / s². -3 m / s², with no significant fluctuations.

[0128] Figures 14-17 This presents the experimental results of hardware-in-the-loop closed-loop control. From... Figure 14-17 As can be seen, the acceleration and velocity curves of the following vehicles 1 and 2 collected in the experiment are highly consistent with the overall evolution of the simulation results. In condition 1, the initial peak acceleration of the two vehicles is about 2.1 m / s² and 1.4 m / s², respectively, which quickly converge to near 0 and fluctuate slightly. The speed increases from 20 m / s to about 31 m / s, and the tracking deviation is within the engineering allowable range within 60 seconds. In condition 2, the acceleration changes sinusoidally, and the peak value is consistent with the simulation. There is no obvious phase lag or amplitude decay. The speed increases from 20 m / s to about 32 m / s, and the tracking deviation is low throughout the process. In condition 3, the acceleration peak is similar to that of condition 1. After 40 seconds, the deceleration peak is about -1.8 m / s². The speed first increases to a constant speed of 32 m / s and then decreases to 25 m / s. The curves almost overlap throughout the process.

Claims

1. A smooth-switching cooperative adaptive cruise control method with dynamic topology, characterized in that, Includes the following steps: Step S1: Design a dynamic topology following information flow architecture to provide scene-adaptive information interaction support for the autonomous vehicle platoon cooperative adaptive cruise control system; the dynamic topology following information flow architecture includes three topology modes: following the preceding vehicle, following the leading vehicle, and following the leading vehicle in both directions, and uses platoon spacing error and communication packet loss rate as two-dimensional decision criteria to achieve adaptive dynamic matching of the topology. Step S2: Construct a topology smoothing switcher based on fuzzy logic to achieve seamless and smooth switching of different information flow topologies through dynamic adjustment of multi-source information weights; the input variables of the topology smoothing switcher are spacing error and communication packet loss rate, and the output variables are the information weight adjustment amounts of the preceding vehicle, the guiding vehicle, and the following vehicle. Step S3: Using a phased progressive training strategy, the queue cooperative cruise controller is constructed and optimized based on the multi-agent proximal policy optimization algorithm. The phased progressive training strategy includes at least a distance tracking phase, a velocity tracking phase, and an acceleration tracking phase.

2. The method for smooth switching cooperative adaptive cruise control with dynamic topology according to claim 1, characterized in that, In step S1, the communication packet loss rate is divided into three levels: 5% is a low packet loss rate, 10% is a medium packet loss rate, and 20% is a high packet loss rate.

3. The method for smooth switching cooperative adaptive cruise control with dynamic topology according to claim 1, characterized in that, In step S2, constructing a topology smoothing switcher based on fuzzy logic specifically includes: dividing the spacing error into three fuzzy subsets: large, medium, and small; dividing the communication packet loss rate into three fuzzy subsets: low, medium, and high; using a triangular membership function to map precise input values ​​to the fuzzy subsets; constructing a fuzzy rule base containing nine core rules; and using the centroid method for defuzzification to convert the fuzzy output set into precise weight adjustment amounts.

4. The method for smooth switching cooperative adaptive cruise control with dynamic topology according to claim 3, characterized in that, The nine core rules are as follows: Rule 1: If the spacing error is medium and the communication packet loss rate is low, then the weight of the preceding vehicle is low, the weight of the guide vehicle is high, and the weight of the following vehicle is low. Rule 2: If the spacing error is medium and the communication packet loss rate is medium, then the weight of the preceding vehicle is low, the weight of the guide vehicle is high, and the weight of the following vehicle is low. Rule 3: If the spacing error is medium and the communication packet loss rate is high, then the weight of the preceding vehicle is medium, the weight of the guide vehicle is medium, and the weight of the following vehicle is low. Rule 4: If the spacing error is large and the communication packet loss rate is low, then the weight of the preceding vehicle is low, the weight of the guide vehicle is low, and the weight of the following vehicle is high. Rule 5: If the spacing error is large and the communication packet loss rate is medium, then the weight of the preceding vehicle is low, the weight of the guide vehicle is medium, and the weight of the following vehicle is medium. Rule 6: If the spacing error is large and the communication packet loss rate is high, then the weight of the preceding vehicle is low, the weight of the guide vehicle is high, and the weight of the following vehicle is low. Rule 7: If the spacing error is small and the communication packet loss rate is low, then the weight of the preceding vehicle is high, the weight of the guide vehicle is low, and the weight of the following vehicle is low. Rule 8: If the spacing error is small and the communication packet loss rate is medium, then the weight of the preceding vehicle is high, the weight of the guide vehicle is low, and the weight of the following vehicle is low. Rule 9: If the spacing error is small and the communication packet loss rate is high, then the weight of the preceding vehicle is high, the weight of the guide vehicle is low, and the weight of the following vehicle is low.

5. The method for smooth switching cooperative adaptive cruise control with dynamic topology according to claim 1, characterized in that, The phased progressive training strategy in step S3 is specifically as follows: Distance tracking phase: A fixed leading vehicle following topology is adopted, and the smooth switching function is disabled. The training objective focuses on maintaining a safe distance. The total reward function for this phase is based on formula (1). for: (1) in, For vehicle spacing tracking reward items, This is a collision hard constraint penalty term. This refers to the weighting coefficient for the vehicle spacing reward item; Speed ​​tracking phase: The topology smooth switching function is enabled, but switching is only allowed between the leading vehicle following and the guide vehicle following topologies. The training objective is to increase speed synchronization while maintaining a safe distance. The total reward function for this phase is based on formula (4): （4） in, This refers to the weighting coefficient of the speed reward item. This is a newly added speed tracking reward item; Acceleration tracking phase: Allows switching between the entire topology, and the training objective further increases the acceleration smoothness. The total reward function for this phase is Equation (6): (6) in, The weighting coefficient for acceleration reward items. This is a newly added acceleration smoothing reward item.

6. The method for smooth switching cooperative adaptive cruise control with dynamic topology according to claim 5, characterized in that, The vehicle spacing tracking reward term in the total reward function of the distance tracking stage adopts formula (2): (2) in, To account for the following vehicle's spacing error, This is the spacing error gain coefficient; The collision hard constraint penalty term adopts formula (3): (3) in, This is the maximum penalty value.

7. The method for smooth switching cooperative adaptive cruise control with dynamic topology according to claim 5, characterized in that, The speed tracking reward term in the total reward function of the speed tracking phase adopts formula (5): (5) in, For speed error, This is the speed error gain coefficient.

8. The method for smooth switching cooperative adaptive cruise control with dynamic topology according to claim 5, characterized in that, The acceleration smoothing reward term in the total reward function of the acceleration tracking phase adopts formula (7): (7) in, To provide real-time acceleration for following the vehicle, The gain coefficient corresponding to the acceleration amplitude. The gain coefficient corresponding to the rate of change of acceleration. This refers to the rate of change of acceleration in real time following the vehicle.

Images