Intelligent driving lane changing decision-making method and system for heavy-duty car

Abstract

The invention belongs to the technical field of vehicle control, and particularly relates to an intelligent driving lane changing decision-making method and system for a heavy-duty car, and the method comprises the steps: defining an automatic driving heavy-duty car as a main car, and recognizing an adjacent car having a potential lane changing conflict with the main car as a competitive car; constructing a probabilistic game model including participation of a main vehicle, a competitive vehicle and environmental disturbance; wherein a state vector and an action set of the main vehicle are defined in the probabilistic game model, and a state transition equation of the main vehicle is established; uncertainties such as vehicle motion states, road curvatures and meteorological conditions are modeled into environmental disturbance in a unified mode, a state evolution law after a main vehicle executes actions is quantitatively described by establishing a state transition equation based on vehicle kinematics, random fluctuation in actual traffic is simulated in combination with Gaussian random disturbance terms, and the real-time performance of the vehicle is improved. And the model can dynamically adapt to complex traffic scenes.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of vehicle control, and particularly relates to an intelligent driving lane-changing decision method and system for heavy-duty vehicles. BACKGROUND

[0002] With the rapid development of advanced driver assistance systems (ADAS) and autonomous driving technologies, vehicle lane-changing decision, as one of the core functions of intelligent driving systems, has an important influence on road traffic efficiency and driving safety. Especially in the field of heavy-duty vehicles, due to the large vehicle mass, high inertia and long braking distance, the reliability and real-time performance of lane-changing decision are more stringent. At present, the mainstream lane-changing decision methods mainly include rule-based methods, machine learning-based methods and game theory-based methods.

[0003] The rule-based method relies on preset heuristic rules or finite state machines, which is effective in simple scenarios, but difficult to cope with the uncertainty of multi-vehicle interaction in complex dynamic traffic environments, and is prone to conservative or delayed decision-making. The machine learning-based method (such as reinforcement learning) generates strategies through data-driven, but has the problem of "black box", poor strategy interpretability, and high requirements for data volume and computing resources, which is insufficient in safety-critical scenarios. The game theory-based method models lane-changing behavior as a multi-agent interaction process, which can describe the strategy game between vehicles, but generally has three shortcomings: first, there is a lack of dynamic adjustment mechanism for the strategy probability distribution and its entropy of traffic participants, which cannot effectively describe the uncertainty; second, the payment function design focuses on a single performance indicator, making it difficult to achieve a balance of safety, efficiency and comfort, especially without adaptive optimization for the characteristics of heavy-duty vehicles such as large load and strong inertia. SUMMARY

[0004] In view of the above shortcomings of the prior art, the application provides an intelligent driving lane-changing decision method and system for heavy-duty vehicles, a terminal and a storage medium.

[0005] In a first aspect, the application provides an intelligent driving lane-changing decision method for heavy-duty vehicles, comprising: S1, defining an autonomous heavy-duty vehicle as a host vehicle, and identifying a neighboring vehicle with potential lane-changing conflict with the host vehicle as a competitor; constructing a probabilistic game model involving the host vehicle, the competitor and environmental disturbance; wherein the state vector of the host vehicle, the action set of the host vehicle and the state transition equation of the host vehicle are defined in the probabilistic game model; S2, defining a safety index based on the prediction result of the state transition equation, defining an efficiency index based on the current speed of the host vehicle, and defining a comfort index based on the current action characteristics in the action set of the host vehicle; S3, calculate the strategy probability distribution of the host vehicle on each action in the action set at the current time, and calculate the strategy entropy based on the probability distribution of the host vehicle; according to the strategy entropy, the dynamic weights of the safety index, the efficiency index and the comfort index are calculated in real time; S4, multiply the safety index, the efficiency index and the comfort index by their corresponding dynamic weights respectively, and sum them up to construct a comprehensive payment function; S5, use the alternating direction multiplier method hierarchical optimization algorithm to iteratively update the strategy probability distribution of the host vehicle, the strategy probability distribution of the competing vehicle and the joint strategy distribution formed by the two, with the goal of minimizing the comprehensive payment function; in the iteration process, a momentum acceleration mechanism is introduced to improve the convergence speed until the preset convergence condition is met; S6, select the action with the highest probability from the final converged strategy probability distribution of the host vehicle as the optimal lane changing strategy and output it for controlling the host vehicle to perform lane changing operation.

[0006] Further improvement of the technical scheme is that step S1 includes: S11, identify a heavy vehicle that is performing an automatic driving task as a decision subject and mark it as a host vehicle, simultaneously perceive the motion state of the vehicle around the host vehicle in real time through a vehicle-mounted sensor, identify the adjacent vehicles located in the adjacent lane of the host vehicle and possibly having spatial trajectory overlap with the host vehicle within a future preset time window as competing vehicles, and model the road curvature change, weather conditions and road adhesion coefficient uncertainty factors as environmental disturbances; S12, in the process of constructing a probabilistic game model involving the host vehicle, the competing vehicles and the environmental disturbances, establish a state vector for the host vehicle ; the state vector at least contains the longitudinal position , the lateral position , the current speed , the heading angle , the road curvature and the real-time rainfall intensity of the host vehicle, and define a candidate action set for the host vehicle, which contains the action of keeping driving in the current lane , the action of performing left lane changing and the action of performing right lane changing ; S13, based on the kinematics of vehicles, establish a state transition equation for the host vehicle, which is used to quantitatively describe the evolution rule of the state vector of the host vehicle from the current time t to the next time t+1 when performing any action in the candidate action set, the state transition equation is composed of a state transition function and a Gaussian random disturbance term , and the expression of the state transition equation is: ; wherein, denotes the state vector of the host vehicle at the next time step predicted after state transition; is the action selected by the host vehicle at time t, ; denotes the action taken by the competing vehicle at time t; is the state transition function, wherein, denotes the longitudinal acceleration generated by the host vehicle when performing action ; denotes the steering angular velocity generated by the host vehicle when performing action ; denotes the time step; and is the position correction term based on the action of the competing vehicle , used to simulate the direct impact of the action of the competing vehicle on the lateral and longitudinal positions of the host vehicle; is the longitudinal acceleration correction term based on the action of the competing vehicle , representing the acceleration adjustment of the host vehicle caused by the action of the competing vehicle; is the steering angular velocity correction term based on the action of the competing vehicle , representing the steering adjustment of the host vehicle triggered by the action of the competing vehicle.

[0007] Further improvements of the technical solution include that step S2 comprises: S21, based on the prediction results of the state transition equation for future multi-step states, calculating the minimum inter-vehicle distance between the host vehicle and surrounding vehicles within a future time window, and constructing a safety index negatively correlated with the risk of collision according to the minimum distance; the safety index is defined as: ; wherein, is the safety index; denotes the minimum Euclidean distance or longitudinal distance between the host vehicle and all adjacent vehicles in the future time window predicted based on the state transition equation, used to represent the potential collision risk; is a preset smoothing factor, ; S22, obtaining the actual driving speed of the host vehicle at the current time, and combining with the road speed limit information to construct an efficiency index reflecting the traffic efficiency; the efficiency index monotonically increases as the speed of the host vehicle approaches the speed limit; the efficiency index is defined as: ; wherein, is the efficiency index; is the highest speed limit value of the current road section; S23, quantifying the smoothness of the driving behavior according to the longitudinal acceleration rate corresponding to the current candidate action in the main vehicle action set, and establishing a comfort index that decays with the acceleration mutation intensifying according to the smoothness of the driving behavior; the comfort index is defined as: ; wherein, is the comfort index; is the longitudinal acceleration rate generated by the main vehicle when performing any candidate action; is a comfort decay coefficient, .

[0008] Further improvements of the technical solution are that step S3 comprises: S31, in the probabilistic game model, for the candidate action set of the main vehicle, calculating the strategy probability distribution of each candidate action of the main vehicle at the current time t , and the expression of the strategy probability distribution ; wherein, represents the estimated value of the action value function of the candidate action selected under the state ; represents the th action in the candidate action set; S32, based on the strategy probability distribution of the main vehicle, calculating the strategy entropy of the main vehicle at time t , and the expression of the strategy entropy ; S33, according to the strategy entropy , real-time calculating the dynamic weights of the safety index, the efficiency index and the comfort index respectively, and the expression of the dynamic weight is: ; wherein, is the dynamic weight of the th index; is the weight adjustment coefficient of the th index.

[0009] Further improvements of the technical solution are that step S4 comprises: S41, obtaining the safety index , the efficiency index and the comfort index at the current state from the definition of step S2; S42, extracting the security dynamic weight from the calculation result in step S3 , the efficiency dynamic weight , and the comfort dynamic weight The three dynamic weight coefficients are normalized values and satisfy ; S43, weighted sum of the three indicators and their corresponding dynamic weights, and superimposed zero mean Gaussian noise term To simulate the random utility fluctuations in the actual traffic environment, a comprehensive payment function is constructed for game optimization; wherein the expression of the comprehensive payment function is: ; Wherein, is a random disturbance term subject to zero mean normal distribution.

[0010] Further improvement of the technical solution is that step S5 includes: S51, initialize the strategy probability distribution of the host vehicle , the strategy probability distribution of the competing vehicle , and the joint strategy probability distribution , wherein the strategy probability distribution of the host vehicle and the competing vehicle is initialized as a uniform distribution, that is For all actions , For all actions , the joint distribution is initialized by tensor product ; At the same time, set the penalty coefficient , momentum coefficient , convergence threshold and maximum iteration number of the alternating direction multiplier method; Initialize momentum variables and to accelerate convergence; wherein, and respectively represent the size of the action set of the host vehicle and the competing vehicle; is a candidate action in the action set of the competing vehicle; S52, fix the strategy probability distribution of the competing vehicle and the joint strategy probability distribution , update the strategy distribution of the host vehicle: ; Wherein, is the updated strategy probability distribution of the host vehicle; denotes mathematical expectation; is the comprehensive payment function defined in step S4; momentum acceleration is applied to calculate the momentum update direction , and the policy probability of the host vehicle is updated as , and respectively represent the momentum accumulation variables of the policy probability distribution of the host vehicle and the competing vehicle at the kth iteration; represents the tensor product operation; S53, calculate the change norm of the policy distribution of the host vehicle between the current iteration and the last iteration , if the change norm is less than the convergence threshold or the number of iterations , stop iteration and output the current policy probability distribution of the host vehicle as the optimal distribution; otherwise, let k=k+1 and return to step S52 to continue iteration.

[0011] Further improvements of the technical solution are as follows: step S6 includes: S61, after the iteration optimization process of step S5 is completed, perform convergence verification on the output policy probability distribution of the host vehicle , and determine whether the output policy probability distribution of the host vehicle meets and satisfies ; wherein represents the final probability value of the host vehicle selecting the ith action; N represents the total number of actions in the action set of the host vehicle; S62, based on the verified policy probability distribution of the host vehicle , select the optimal lane changing action from the action set A of the host vehicle by the probability maximization principle : ; wherein, the argmax operation represents selecting the action that makes the probability value reach the maximum as the optimal decision output; S63, convert the optimal lane changing strategy determined in step S62 into a vehicle control instruction set , including the target lane number , the expected longitudinal acceleration , and the expected yaw rate , and output the generated vehicle control instruction to the drive-by-wire execution system of the vehicle, and the vehicle control instruction is represented as: ; wherein, is a policy-instruction conversion function.

[0012] In a second aspect, the present application provides an intelligent driving lane changing decision system for heavy trucks, comprising:​ A probabilistic game model construction module is configured to define the autonomous heavy vehicle as a host vehicle and identify a neighboring vehicle having a potential lane-changing conflict with the host vehicle as a competitor vehicle, construct a probabilistic game model involving the host vehicle, the competitor vehicle and environmental disturbance, define a state vector of the host vehicle, an action set of the host vehicle and a state transition equation of the host vehicle in the probabilistic game model. An index definition module is configured to define a safety index based on a prediction result of the state transition equation, define an efficiency index based on a current speed of the host vehicle, and define a comfort index based on a current action characteristic in the action set of the host vehicle. A strategy entropy calculation module is configured to calculate a strategy probability distribution of each action in the action set of the host vehicle at a current time and calculate a strategy entropy based on the strategy probability distribution of the host vehicle, and calculate dynamic weights of the safety index, the efficiency index and the comfort index based on the strategy entropy. A comprehensive payment function construction module is configured to multiply the safety index, the efficiency index and the comfort index by the corresponding dynamic weights respectively and sum the results to construct a comprehensive payment function. A model iteration module is configured to use an alternating direction multiplier method hierarchical optimization algorithm to iteratively update a strategy probability distribution of the host vehicle, a strategy probability distribution of the competitor vehicle and a joint strategy distribution formed by the two strategy probability distributions, with the objective of minimizing the comprehensive payment function, and introduce a momentum acceleration mechanism in the iteration process to improve the convergence speed until a preset convergence condition is met. A lane-changing strategy output module is configured to select an action having the highest probability from the finally converged strategy probability distribution of the host vehicle as an optimal lane-changing strategy and output the optimal lane-changing strategy for controlling the host vehicle to perform a lane-changing operation.

[0013] In a third aspect, the present application provides a terminal, comprising: a processor and a memory, the memory is configured to store a computer program, the processor is configured to call and run the computer program from the memory, so that the terminal executes the method of the terminal described above.

[0014] In a fourth aspect, the present application provides a computer storage medium, which stores instructions, when the instructions are run on a computer, the computer executes the method described in the above aspects.

[0015] The present application has the following advantages: The application unifies the uncertainties such as vehicle motion state, road curvature, and weather conditions into environmental disturbances by constructing a probabilistic game model including a host vehicle, a competing vehicle, and environmental disturbances, and defines a state vector (such as longitudinal / lateral position, speed, heading angle, etc.) of the host vehicle and an action set (lane keeping, left / right lane changing). By establishing a state transition equation based on vehicle kinematics, the state evolution law after the host vehicle performs an action is quantitatively described, and a Gaussian random disturbance term is combined to simulate random fluctuations in actual traffic, so that the model can dynamically adapt to complex traffic scenarios. For example, in rainy weather or on a curved road section, the environmental disturbance term can automatically adjust the state transition prediction result to avoid decision bias caused by static assumptions of the model, and significantly improve the robustness of the decision.

[0016] The application quantifies the degree of certainty of the host vehicle action selection by introducing policy entropy, and calculates the dynamic weights of safety, efficiency, and comfort indicators in real time based on policy entropy. For example: Safety indicator: a collision risk model is constructed by predicting the minimum inter-vehicle distance between the host vehicle and the competing vehicle, and the weight increases with the risk, ensuring that safety is prioritized in high-risk scenarios; Efficiency indicator: combined with the current speed of the host vehicle and the road speed limit information, the weight increases as the vehicle speed approaches the speed limit, improving the traffic efficiency on high-speed smooth road sections; Comfort indicator: based on the longitudinal acceleration change rate to quantify driving smoothness, the weight decreases as the acceleration mutation intensifies, suppressing the impact of sudden acceleration / sudden deceleration on passenger comfort.

[0017] Through dynamic weight adjustment, the application realizes adaptive balancing of multiple target indicators, especially for heavy vehicles with large load and strong inertia, avoiding suboptimal decisions caused by excessive optimization of a single indicator.

[0018] The application uses an alternating direction multiplier method (ADMM) hierarchical optimization algorithm to update the policy probability distribution of the host vehicle and the competing vehicle and the joint policy distribution iteratively, with the comprehensive payment function as the target. By introducing a momentum acceleration mechanism, the momentum accumulation variable of historical iteration information is used to correct the current update direction, significantly improving the convergence speed. For example, in complex traffic scenarios, the traditional method may require hundreds of iterations to converge, while the application can reduce the number of iterations to tens of times through momentum acceleration, ensuring that the decision is completed within milliseconds, meeting the requirements of real-time control of heavy vehicles.

[0019] The application constructs a comprehensive payment function by weighting and summing the safety, efficiency, and comfort indicators, and adding a zero-mean Gaussian noise term to simulate random utility fluctuations in actual traffic. This design allows the decision-making process to consider both deterministic optimization and randomness adaptation, for example: In a deterministic scenario, the payment function realizes multi-objective balancing through weight adjustment; In the random disturbance scene (for example, a competing car suddenly changes lanes), the noise term can trigger the dynamic adjustment of the strategy probability distribution, avoiding the decision failure caused by the model overfitting the prediction results.

[0020] In addition, the normalization dynamic weight design of the payment function ensures the consistency of the dimension of each indicator, avoiding the optimization deviation caused by the too large difference in the numerical value of the indicator. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0022] Figure 1 The schematic flowchart of the method of one embodiment of the present application.

[0023] Figure 2 The schematic block diagram of the system of one embodiment of the present application.

[0024] Figure 3 The structural schematic diagram of a terminal provided by an embodiment of the present application. DETAILED DESCRIPTION

[0025] In order to make the objectives, characteristics and advantages of the present application more obvious and easy to understand, the technical solutions in the present application will be described clearly and completely in combination with the drawings in the specific embodiments. Obviously, the following described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terms used in the specification of the present application are only for the purpose of describing the specific embodiments and are not intended to limit the present application.

[0027] Figure 1 The schematic flowchart of the method of one embodiment of the present application provides an intelligent driving lane changing decision method for a heavy truck. In the flowchart, Figure 1 The execution subject can be an intelligent driving lane changing decision system for a heavy truck. According to different needs, the order of the steps in the flowchart can be changed, and some can be omitted.

[0028] As Figure 1 shown, the method comprises: S1, define the autonomous heavy vehicle as a host vehicle, and identify the adjacent vehicle with potential lane-changing conflict with the host vehicle as a rival vehicle; construct a probabilistic game model involving the host vehicle, the rival vehicle and environmental disturbance; wherein the state vector of the host vehicle, the action set of the host vehicle and the state transition equation of the host vehicle are defined in the probabilistic game model; S2, define the safety index based on the prediction result of the state transition equation, define the efficiency index based on the current speed of the host vehicle, and define the comfort index based on the current action characteristics in the action set of the host vehicle; S3, calculate the strategy probability distribution of each action in the action set of the host vehicle at the current time, and calculate the strategy entropy based on the probability distribution of the host vehicle; according to the strategy entropy, the dynamic weights of the safety index, the efficiency index and the comfort index are calculated in real time; S4, multiply the safety index, the efficiency index and the comfort index by the corresponding dynamic weight respectively, and sum them up to construct a comprehensive payment function; S5, use the alternating direction multiplier method hierarchical optimization algorithm to minimize the comprehensive payment function as the target, iteratively update the strategy probability distribution of the host vehicle, the strategy probability distribution of the rival vehicle and the joint strategy distribution formed by the two; in the iteration process, the momentum acceleration mechanism is introduced to improve the convergence speed until the preset convergence condition is met; S6, select the action with the highest probability from the finally converged strategy probability distribution of the host vehicle as the optimal lane-changing strategy and output it for controlling the host vehicle to perform the lane-changing operation.

[0029] In order to facilitate the understanding of the present application, the principle of the intelligent driving lane-changing decision method of the heavy vehicle of the present application is described further in combination with the process of intelligent driving lane-changing decision for the heavy vehicle in the embodiment.

[0030] Firstly, step S1 includes: S11, identify the heavy vehicle performing the automatic driving task as the decision subject and mark it as the host vehicle, simultaneously perceive the motion state of the vehicle around the host vehicle in real time through the vehicle-mounted sensor, identify the adjacent vehicle in the adjacent lane of the host vehicle and possibly having spatial trajectory overlap with the host vehicle within the future preset time window as the rival vehicle, and model the road curvature change, weather condition and road adhesion coefficient uncertainty factors as the environmental disturbance; S12, in the process of constructing the probabilistic game model involving the host vehicle, the rival vehicle and the environmental disturbance, the state vector of the host vehicle is established ; the state vector at least contains the longitudinal position , the lateral position , the current speed , the heading angle road curvature and real-time rainfall intensity and define a candidate action set for the host vehicle the candidate action set contains a keep current lane driving action a left lane changing action and a right lane changing action ; S13, based on the kinematics of the vehicle, a state transition equation is established for the host vehicle, the state transition equation is used to quantitatively describe the evolution rule of the state vector of the host vehicle from the current time t to the next time t+1 when performing any action in the candidate action set, the state transition equation is composed of a state transition function and a Gaussian random disturbance term , and the expression of the state transition equation is: ; wherein, represents the state vector of the host vehicle at the next time t+1 predicted after state transition; is an action selected by the host vehicle at time t, ; represents the action taken by the competing vehicle at time t; is a state transition function, wherein, represents the longitudinal acceleration of the host vehicle when performing the action ; represents the steering angular velocity of the host vehicle when performing the action ; represents the time step; and are position correction terms based on the action of the competing vehicle, used to simulate the direct influence of the action of the competing vehicle on the lateral and longitudinal positions of the host vehicle; is a longitudinal acceleration correction term based on the action of the competing vehicle, representing the acceleration adjustment of the host vehicle caused by the action of the competing vehicle; is a steering angular velocity correction term based on the action of the competing vehicle, representing the steering adjustment of the host vehicle triggered by the action of the competing vehicle.

[0031] S11, participant identification and environment modeling: In this step, the surrounding environment information of the host vehicle is collected in real time through the vehicle-mounted sensor system (including but not limited to lidar, millimeter wave radar, camera, GPS / IMU combined navigation system). The specific implementation process is as follows: Main vehicle identification: the HOWO series heavy truck performing the automatic driving task is taken as the main vehicle, whose whole vehicle mass is ≥18 tons and the vehicle length is ≥12 meters. The main vehicle is equipped with an automatic driving computing platform based on NVIDIA DRIVE AGX Orin.

[0032] Competitive vehicle identification: a target detection algorithm based on the improved YOLOv5 is adopted to detect the vehicles within a range of ±60 meters around the main vehicle in real time. The motion trajectory of the vehicle within the next 5 seconds is predicted through the Kalman filtering algorithm, and the vehicle meeting the following conditions is marked as the competitive vehicle: ; wherein, and represent the predicted positions of the main vehicle and the surrounding vehicle at time, is a safety distance threshold, and the value is 8 meters.

[0033] Environment disturbance modeling: the real-time rainfall intensity (unit: ) is obtained through the vehicle-mounted weather station, the road curvature (unit: ) is obtained through the high-precision map, and the road adhesion coefficient is obtained through the road surface recognition algorithm. These three are modeled as an environmental disturbance term , and the covariance matrix thereof is: ; wherein, the variances of the components are calibrated according to the measured data: , , , , , .

[0034] S12, state vector and action set definition: State vector construction: the main vehicle state vector is represented by a six-dimensional vector, and the specific definition is as follows: ; wherein, is the longitudinal position of the main vehicle in the UTM coordinate system, which is obtained through the GPS positioning system; is the lateral position of the main vehicle in the UTM coordinate system, which is obtained through the GPS positioning system; is the current speed of the main vehicle, which is calculated through the fusion of the wheel speed sensor and the IMU; is the heading angle of the main vehicle, which is measured through the IMU; is the road curvature, which is obtained through the high-precision map query; For real-time rainfall intensity, measured by vehicle-mounted rain sensor.

[0035] S13, longitudinal position correction term wherein, is the longitudinal coupling coefficient (value range 0.1-0.5), obtained by real vehicle data calibration; is the longitudinal speed of the competing vehicle at time t (unit: m / s), measured by millimeter wave radar; is the current longitudinal speed of the host vehicle (unit: m / s); is the longitudinal distance between the two vehicles (unit: m), calculated by GPS differential positioning; is the safe following distance (unit: m), the calculation formula is wherein, is the preset headway, is the basic safety distance.

[0036] Lateral position correction term : wherein, is the lateral coupling coefficient (value range 0.05-0.2), calibrated by curve test; is the lateral action component of the competing vehicle (unit: m / s²), obtained by V2X communication or visual sensor; is the lateral position of the competing vehicle (unit: m); is the current lateral position of the host vehicle (unit: m); is the road speed limit value (unit: m / s).

[0037] Longitudinal acceleration correction term : wherein, is the acceleration adjustment gain (value range 0.3-1.0), adjusted based on vehicle mass (larger value for heavy trucks); is the relative speed (unit: m / s), , represents the real-time speed of the competing vehicle at time t; is the speed difference standardization parameter; is the real-time minimum vehicle distance (unit: m); is the critical safety distance, set according to GB 7258 standard.

[0038] Steering angular velocity correction term : ​​​​ ; is the steering regulation coefficient (value range 0.01-0.1 rad / s), which is positively correlated with the road adhesion coefficient; is the competitive vehicle steering action (unit: rad / s); is the current road curvature (unit: 1 / m); is the curvature smoothing factor, to avoid division by zero error; is the heading angle difference (unit: rad), , represents the real-time heading angle of the competitive vehicle at time t; is the heading angle standard deviation, .

[0039] The application identifies the competitive vehicle through multi-sensor fusion and trajectory prediction algorithm, avoids missed or mistaken judgment, and quantizes and encapsulates the road, weather and road conditions, solves the problem of fuzzy modeling of uncertain factors in traditional methods, and adapts to the characteristics of high sensitivity of heavy trucks to the environment.

[0040] The state vector covers the core parameters required for heavy truck lane changing decision, the quantization accuracy and acquisition method of each dimension parameter meet the engineering practice, and the kinetic parameters of the action set are optimized according to the characteristics of large inertia and long braking distance of heavy trucks, so as to avoid the lane changing risk caused by unreasonable action parameters.

[0041] Based on the kinematic principle, the state transition equation is constructed, combined with the Gaussian random disturbance term to simulate random factors in the actual scene, and the prediction result can accurately reflect the state evolution of the host vehicle under different actions and environments, providing reliable basic data for subsequent safety and efficiency index calculation.

[0042] Secondly, step S2 comprises: S21, based on the prediction result of the future multi-step state of the state transition equation, the minimum inter-vehicle distance of the host vehicle and the surrounding vehicles in the future time window is calculated, and a safety index negatively correlated with the collision risk is constructed according to the minimum distance; the safety index is defined as: ; wherein, is the safety index; represents the minimum Euclidean distance or longitudinal distance between the host vehicle and all adjacent vehicles in the future time window predicted based on the state transition equation, which is used to represent the potential collision risk; is a preset smoothing factor, ; S22, the actual driving speed of the host vehicle at the current time is obtained, and the efficiency index reflecting the traffic efficiency is constructed in combination with the road speed limit information, the efficiency index monotonically increases as the speed of the host vehicle approaches the speed limit; the efficiency index is defined as: ; wherein, is the efficiency index; is the highest speed limit value of the current road section, the closer the current speed of the host vehicle approaches the highest speed limit value, the higher the score of the efficiency index; S23, the smoothness of the driving behavior is quantified according to the longitudinal acceleration change rate corresponding to the current candidate action in the host vehicle action set, and the comfort index decaying with the acceleration mutation intensifying is established according to the smoothness of the driving behavior; the comfort index is defined as: ; wherein, is the comfort index; is the longitudinal acceleration change rate generated by the host vehicle when executing any candidate action; is a comfort decay coefficient, .

[0043] S21, safety index construction: The safety index quantifies the collision risk by predicting the future trajectory and calculating the minimum distance, and the specific implementation is as follows: Trajectory prediction and distance calculation: Based on the state transition equation established in S1, the trajectories of the host vehicle and each competitor in the future time window are predicted, and the sampling interval ; the relative positions of the host vehicle and the competitor at each sampling time are calculated: ; The minimum distance of all sampling time is taken as : When , it is considered as high risk, triggering the emergency braking auxiliary system.

[0044] S22, efficiency index construction: The efficiency index evaluates the traffic efficiency through the ratio of the vehicle speed to the road speed limit: Speed data processing: The host vehicle speed is obtained through the CAN bus, and moving average filtering (window size 5) is used to eliminate noise; the road speed limit is dynamically obtained through a high-precision map, supporting different lane differentiated speed limits. When , take To avoid encouraging speeding, a weighting coefficient of 0.6 is introduced for special road sections such as ramps and toll stations to appropriately reduce efficiency requirements.

[0045] S23, Construction of Comfort Indicators: Comfort metrics quantify ride comfort using jerk (rate of change of acceleration): Calculation of jerk: Based on action Corresponding longitudinal acceleration Calculate jerk: ; A first-order low-pass filter (cutoff frequency 2Hz) is used for smoothing to eliminate high-frequency noise.

[0046] Next, step S3 includes: S31. In a probabilistic game model, the set of candidate actions for the main vehicle. Calculate the strategy probability distribution of the master vehicle for each candidate action at the current time t. Strategy probability distribution The expression is: ; in, Indicates the state Next, select candidate actions The estimated value of the action value function; Represents the first action in the candidate action set. One action; S32, Probability Distribution of Policies Based on the Master Vehicle Calculate the strategy entropy of the master vehicle at time t. Policy entropy is used to quantify the degree of uncertainty in the current lane-changing decision of the master vehicle. The expression is: ; Among them, the larger the strategy entropy value, the higher the uncertainty of the master vehicle's choice of lane changing action, and the probability of each action being selected tends to be evenly distributed. S33, Based on strategy entropy The dynamic weights of safety, efficiency, and comfort indicators are calculated in real time. The expression for the dynamic weights is as follows: ; in, is the dynamic weight of the i-th indicator; Let be the weighting adjustment coefficient for the i-th indicator. .

[0047] S31. Calculation of strategy probability distribution: The policy probability distribution converts the action value function into a probability distribution through a softmax function, and is implemented as follows: The action value function is estimated as: The value function of each action is estimated based on a deep Q network (DQN) , and the network structure is as follows: Input layer: state vector with 6 dimensions (position, velocity, heading angle, curvature, rainfall intensity); Hidden layer: 3 fully connected layers with 128, 64, and 32 neurons, respectively, using a ReLU activation function; Output layer: 3 neurons, corresponding to , , .

[0048] The action value function is updated using time-difference learning: ; wherein is the learning rate, is the discount factor.

[0049] Numerical stability processing: To prevent overflow of exponential operation, the maximum value reduction technique is used: .

[0050] In addition, step S4 includes: S41, obtaining the safety index , efficiency index , and comfort index under the current state from the definition in step S2; S42, extracting the safety dynamic weight , efficiency dynamic weight , and comfort dynamic weight from the calculation results in step S3, the three dynamic weight coefficients are normalized values and satisfy ; S43, weighted sum of the three indexes and their corresponding dynamic weights, and superimpose a zero-mean Gaussian noise term to simulate the random utility fluctuations in the actual traffic environment, and construct a comprehensive payment function for game optimization; wherein the expression of the comprehensive payment function is: ; wherein is a random disturbance term subject to a zero-mean normal distribution.

[0051] S41, obtaining a multi-target performance index: Index data collection and preprocessing: Safety index acquisition: from S2 calculation Read the current value, data format is floating point number, range [0, 1]. When , When , (maximum safety risk).

[0052] Efficiency index acquisition: real-time reading, ensuring data validity check: Check whether it is within a reasonable range (0-120km / h); Check whether it matches the current road type; When , .

[0053] Comfort index acquisition: read from S2 calculation results and perform smoothing processing: .

[0054] Data synchronization mechanism: Three indexes are synchronized by timestamp to ensure the same state data at the same time; Set data validity period , data beyond the validity period is considered invalid.

[0055] Then, step S5 includes: S51, initializing the strategy probability distribution of the host vehicle , the strategy probability distribution of the competing vehicle and the joint strategy probability distribution , wherein the strategy probability distribution of the host vehicle and the competing vehicle is initialized as a uniform distribution, i.e. for all actions , for all actions , the joint distribution is initialized by tensor product ; at the same time, set the penalty coefficient , momentum coefficient , convergence threshold and maximum iteration number of the alternating direction multiplier method; initialize momentum variables and for accelerating convergence; wherein and represent the size of the action set of the host vehicle and the competing vehicle respectively; Set of candidate actions in the competition car; S52, update the strategy probability distribution of the competition car and the joint strategy probability distribution , update the strategy distribution of the host car: ; where, is the updated strategy probability distribution of the host car; denotes the mathematical expectation; is the comprehensive payoff function defined in step S4; apply momentum acceleration to calculate the momentum update direction , and update the strategy probability of the host car as , and respectively represent the momentum accumulation variables of the strategy probability distributions of the host car and the competition car at the kth iteration; denotes the tensor product operation; S53, calculate the change norm of the strategy distribution of the host car between the current iteration and the last iteration If the change norm is less than the convergence threshold or the number of iterations , stop the iteration, and output the current strategy probability distribution of the host car as the optimal distribution; otherwise, let k=k+1 and return to step S52 to continue iteration.

[0056] In addition, the method also includes updating the strategy probability distribution of the competition car: Fix the host car distribution and the joint distribution, and update the strategy probability distribution of the competition car by solving the following optimization problem: ; where, is the updated strategy probability distribution of the competition car; then apply momentum acceleration to calculate the momentum update direction , and update the strategy probability distribution of the competition car as .

[0057] Update the joint strategy distribution: update the joint distribution according to the updated host car and competition car distributions as: .

[0058] S51, initialize the strategy distribution and parameter settings: Strategy probability distribution initialization: Host car strategy probability distribution initialization: assuming that the host car action set A contains 3 actions (keep lane, left lane change, right lane change), the initial distribution is a uniform distribution: ​ ; in This indicates the size of the set of actions of the main vehicle.

[0059] Initialization of the probability distribution of competing vehicle strategies: Suppose that the set of competing vehicle actions B contains two actions (lane keeping and lane changing), then the initial distribution is as follows: ; in This indicates the size of the set of actions of the main vehicle.

[0060] Initialization of joint policy probability distribution: Calculate the initial joint distribution using tensor product: ; The specific calculation is as follows: This generates 6 joint probability values.

[0061] Momentum variable initialization: Main vehicle momentum variables: ; Momentum variables of competing vehicles: .

[0062] Calculation of the expected value of the payoff function: ; in The comprehensive payment function defined for S4.

[0063] Finally, step S6 includes: S61. After the iterative optimization process in step S5 is completed, the output master vehicle strategy probability distribution is... Perform convergence verification and determine whether the output master vehicle strategy probability distribution satisfies the following conditions. And satisfy ;in, This represents the final probability value of the master vehicle selecting the i-th action; N represents the total number of actions in the master vehicle's action set; S62, Based on the verified master vehicle strategy probability distribution The optimal lane-changing maneuver is selected from the set of master vehicle actions A using the principle of maximizing probability. : ; The argmax operation represents selecting the probability value. To achieve the maximum movement As the optimal decision output; S63. Apply the optimal lane-changing strategy determined in step S62. Convert to vehicle control instruction set Including the target lane number , expected longitudinal acceleration and expected yaw angular velocity and outputting the generated vehicle control instructions to a drive-by-wire execution system of the vehicle, the vehicle control instructions being represented as: ; wherein, is a strategy-instruction conversion function.

[0064] In some embodiments, the intelligent driving lane-changing decision system 200 for heavy-duty vehicles can include a plurality of functional modules composed of computer program segments. The computer programs of each program segment in the intelligent driving lane-changing decision system 200 for heavy-duty vehicles can be stored in the memory of a computer device and executed by at least one processor to perform the functions of the intelligent driving lane-changing decision system 200 for heavy-duty vehicles (see detailed description below). Figure 1

[0065] In this embodiment, the intelligent driving lane-changing decision system 200 for heavy-duty vehicles can be divided into a plurality of functional modules according to the functions it performs, as shown in Figure 2 The functional modules can include a probabilistic game model construction module 210, an index definition module 220, a strategy entropy calculation module 230, a comprehensive payment function construction module 240, a model iteration module 250, and a lane-changing strategy output module 260. The module referred to in the present application refers to a series of computer program segments that can be executed by at least one processor and can complete a fixed function, which are stored in the memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0066] ​Specifically, the probabilistic game model construction module 210 is configured to define the autonomous heavy vehicle as a host vehicle, and identify a neighboring vehicle having a potential lane-changing conflict with the host vehicle as a competitor vehicle; construct a probabilistic game model involving the host vehicle, the competitor vehicle and environmental disturbance; wherein the state vector of the host vehicle, the action set of the host vehicle and the state transition equation of the host vehicle are defined in the probabilistic game model; the index definition module 220 is configured to define a safety index based on the prediction result of the state transition equation, define an efficiency index based on the current speed of the host vehicle, and define a comfort index based on the current action characteristics in the action set of the host vehicle; the strategy entropy calculation module 230 is configured to calculate the strategy probability distribution of each action in the action set of the host vehicle at the current time, and calculate the strategy entropy based on the probability distribution of the host vehicle; according to the strategy entropy, the dynamic weights of the safety index, the efficiency index and the comfort index are calculated in real time; the comprehensive payoff function construction module 240 is configured to multiply the safety index, the efficiency index and the comfort index by the corresponding dynamic weights respectively, and sum them to construct a comprehensive payoff function; the model iteration module 250 is configured to use an alternating direction multiplier method hierarchical optimization algorithm to iteratively update the strategy probability distribution of the host vehicle, the strategy probability distribution of the competitor vehicle and the joint strategy distribution formed by the two, with the goal of minimizing the comprehensive payoff function; a momentum acceleration mechanism is introduced in the iteration process to improve the convergence speed until the preset convergence condition is met; the lane-changing strategy output module 260 is configured to select the action with the highest probability from the finally converged strategy probability distribution of the host vehicle as the optimal lane-changing strategy and output it for controlling the host vehicle to perform the lane-changing operation.

[0067] Figure 3 A structure schematic diagram of a terminal 300 is provided for the embodiments of the present application, and the terminal 300 can be used to execute the intelligent driving lane-changing decision method for heavy vehicles provided by the embodiments of the present application.

[0068] The terminal 300 can include a processor 310, a memory 320 and a communication module 330. These components communicate through one or more buses, and those skilled in the art can understand that the structure of the server shown in the figure does not constitute a limitation on the present application, which can be a bus structure or a star structure, and can include more or fewer components than shown in the figure, or combine some components, or different component arrangements.

[0069] The memory 320 can be used to store the execution instructions of the processor 310, and the memory 320 can be implemented by any type of volatile or non-volatile storage terminal or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk. When the execution instructions in the memory 320 are executed by the processor 310, the terminal 300 can execute part or all of the steps in the following method embodiments.

[0070] The processor 310 is the control center of the storage terminal, connects various parts of the entire electronic terminal through various interfaces and lines, executes the software programs and / or modules stored in the memory 320, and calls the data stored in the memory, so as to execute various functions of the electronic terminal and / or process data. The processor can be composed of an integrated circuit (IC), for example, can be composed of a single packaged IC, or can be composed of a plurality of packaged ICs with the same function or different functions. For example, the processor 310 can only include a central processing unit (CPU). In the embodiments of the present application, the CPU can be a single operation core or can include multiple operation cores.

[0071] The communication module 330 is used to establish a communication channel, so that the storage terminal can communicate with other terminals. The communication module 330 receives user data sent by other terminals or sends user data to other terminals.

[0072] The present application also provides a computer storage medium, wherein the computer storage medium can store a program, and the program can include part or all of the steps in the embodiments provided by the present application when executed. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0073] Those skilled in the art can clearly understand that the technical solutions in the embodiments of the present application can be implemented by means of software plus necessary universal hardware platforms. Based on such an understanding, the technical solutions in the embodiments of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a USB flash disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and the like, and includes a plurality of instructions for causing a computer terminal (which can be a personal computer, a server, or a second terminal, a network terminal, or the like) to execute all or part of the steps of the methods described in the embodiments of the present application.

[0074] In the present specification, the same or similar parts among various embodiments can be referred to each other. In particular, for the terminal embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description in the method embodiments.

[0075] In several embodiments provided in the present application, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiments described above are merely schematic, for example, the division of the modules is only a logical function division, and actual implementation can have another division manner, for example, a plurality of modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed modules can be indirect coupling or communication connection through some interfaces, systems or modules, and can be electrical, mechanical or other forms.

[0076] The modules described as separate components can or can not be physically separated, and the components displayed as modules can or can not be physical modules, that is, they can be located in one place, or can be distributed on a plurality of network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments.

[0077] In addition, the functional modules in each of the embodiments of the present application can be integrated in one processing module, or each module can exist physically independently, or two or more modules can be integrated in one module.

[0078] Although the present application has been described in detail with reference to the preferred embodiments, it should be understood that the application is not limited to those preferred embodiments. Various modifications and equivalents can be made by those skilled in the art without departing from the spirit and scope of the application. Any and all modifications and equivalents are intended to be included within the scope of the present application.

Claims

1. An intelligent driving lane-changing decision method for heavy-duty vehicles, characterized in that, The application comprises the following steps: S1, defining an automatic driving heavy truck as a host vehicle, and identifying an adjacent vehicle with a potential lane-changing conflict with the host vehicle as a competitor; a probabilistic game model involving the host vehicle, the competitor and environmental disturbances is constructed; wherein, the state vector of the host vehicle, the action set of the host vehicle and the state transition equation of the host vehicle are defined in the probabilistic game model; S2, defining a safety index based on the prediction result of the state transition equation, defining an efficiency index based on the current speed of the host vehicle, and defining a comfort index based on the current action characteristics in the action set of the host vehicle; S3, calculating the strategy probability distribution of each action in the action set of the host vehicle at the current time, and calculating the strategy entropy based on the probability distribution of the host vehicle; according to the strategy entropy, the dynamic weights of the safety index, the efficiency index and the comfort index are calculated in real time; S4, multiplying the safety index, the efficiency index and the comfort index by their corresponding dynamic weights and summing them up to construct a comprehensive payment function; S5, using an alternating direction multiplier method hierarchical optimization algorithm to minimize the comprehensive payment function as the target, iteratively updating the strategy probability distribution of the host vehicle, the strategy probability distribution of the competitor and the joint strategy distribution formed by the two; in the iteration process, a momentum acceleration mechanism is introduced to improve the convergence speed until the preset convergence condition is met; S6, selecting the action with the highest probability from the finally converged strategy probability distribution of the host vehicle as the optimal lane-changing strategy and outputting it for controlling the host vehicle to perform lane-changing operation.

2. The intelligent driving lane-changing decision method for heavy-duty vehicles according to claim 1, characterized in that, Step S1 comprises: S11, identifying a heavy truck performing an automatic driving task as a decision subject and recording it as a host vehicle, simultaneously sensing the motion state of vehicles around the host vehicle in real time through a vehicle-mounted sensor, identifying an adjacent vehicle located in the adjacent lane of the host vehicle and possibly having a spatial trajectory overlap with the host vehicle within a future preset time window as a competitor, and modeling road curvature changes, weather conditions and road adhesion coefficient uncertainties as environmental disturbances; S12, in the process of constructing a probabilistic game model involving the main vehicle, the competing vehicle and the environmental disturbance, a state vector is established for the main vehicle ; the state vector at least contains the longitudinal position , the lateral position , the current vehicle speed , the heading angle , the road curvature and the real-time rainfall intensity of the main vehicle, and a candidate action set is defined for the main vehicle , the candidate action set contains the action of keeping driving in the current lane , the action of performing left lane changing and the action of performing right lane changing ; S13. Based on the kinematic principle of the vehicle, a state transition equation for the host vehicle is established, the state transition equation being used to quantitatively describe the host vehicle in executing the candidate action set at any action, evolution law of a state vector thereof from a current time t to a next time t+1, the state transition equation being composed of a state transition function and a Gaussian random disturbance term , and an expression of the state transition equation is as follows: ； wherein, denotes the state vector of the host vehicle at the next time instant predicted after the state transition; is a function of the state of the host vehicle at time instant t, ; denotes the action taken by the competing vehicle at time instant t; is a state transition function, wherein, denotes the longitudinal acceleration generated by the host vehicle when performing the action ; denotes the steering angular velocity generated by the host vehicle when performing the action ; denotes the time step; and is a position correction term based on the action of the competing vehicle; is a longitudinal acceleration correction term based on the action of the competing vehicle; is a steering angular velocity correction term based on the action of the competing vehicle.

3. The intelligent driving lane-changing decision method for heavy-duty vehicles according to claim 2, characterized in that, Step S2 comprises: S21, based on the prediction result of the state transition equation for future multi-step states, calculating the minimum distance between the host vehicle and the surrounding vehicles within the future time window, and constructing a safety index negatively correlated with the risk of collision according to the minimum distance; the safety index is defined as: ； wherein, is a safety index; denotes the minimum Euclidean distance or longitudinal gap between the host vehicle and all neighboring vehicles in the future time window predicted based on the state transition equation, for characterizing the potential collision risk; is a preset smoothing factor, ; S22, obtaining the actual driving speed of the host vehicle at the current time, and combining with the road speed limit information to construct an efficiency index reflecting the passing efficiency; the efficiency index monotonically increases as the speed of the host vehicle approaches the speed limit; the efficiency index is defined as: ； wherein, is an efficiency index; is a maximum speed limit value for the current road segment; S23, quantifying the smoothness of the driving behavior according to the longitudinal acceleration change rate corresponding to the current candidate action in the action set of the host vehicle, and establishing a comfort index that decays with the acceleration mutation intensifying; the comfort index is defined as: ； wherein, is a comfort index; is a longitudinal acceleration rate of change produced by the host vehicle when performing any of the candidate actions; is a comfort decay coefficient, .

4. The intelligent driving lane-changing decision method for heavy-duty vehicles according to claim 3, characterized in that, Step S3 comprises: S31、In the probabilistic game model, for the candidate action set of the host vehicle , calculate the strategy probability distribution of each candidate action of the host vehicle at the current time t , the expression of the strategy probability distribution is: ； wherein, represents an estimated value of an action value function that selects a candidate action under a state ; represents an i-th action in a candidate action set ; S32, a strategy probability distribution of the host vehicle , calculating a strategy entropy of the host vehicle at time t , the strategy entropy The expression is: ； S33、According to the policy entropy , real-time calculation of the dynamic weight of the security index, the efficiency index and the comfort index, the expression of the dynamic weight is: ； wherein, is a dynamic weight for the first index; is a weight adjustment coefficient for the first index, .

5. The intelligent driving lane-changing decision method for heavy-duty vehicles according to claim 4, characterized in that, Step S4 comprises: S41, obtaining the safety index in the current state from the definition in step S2 , efficiency index , and comfort index ; S42, extracting the security dynamic weight from the calculation result in step S3 , the efficiency dynamic weight , and the comfort dynamic weight The three dynamic weight coefficients are normalized values and satisfy ; S43. Calculate the weighted sum of the three indicators and their corresponding dynamic weights, and then add a zero-mean Gaussian noise term. To simulate stochastic utility fluctuations in a real traffic environment, a comprehensive payoff function is constructed for game optimization; whereby the comprehensive payoff function... The expression is: ； wherein is a random disturbance term that is subject to a zero-mean normal distribution.

6. The intelligent driving lane-changing decision method for heavy-duty vehicles according to claim 5, characterized in that, Step S5 comprises: S51, initialize strategy probability distribution of the host vehicle S52, initialize strategy probability distribution of the competing vehicle S53, initialize joint strategy probability distribution where the strategy probability distributions of the host vehicle and the competing vehicle are initialized as uniform distributions, i.e. for all actions , for all actions , the joint distribution is initialized by tensor product ; the penalty coefficient , the momentum coefficient , the convergence threshold and the maximum iteration number of the alternating direction multiplier method are set; the momentum variables and are initialized to accelerate the convergence; wherein, and respectively represent the size of the action set of the host vehicle and the competing vehicle; is a candidate action in the action set of the competing vehicle; S52, update strategy probability distribution of the host vehicle and joint strategy probability distribution , update strategy distribution of the host vehicle: ； wherein, is the updated master car strategy probability distribution; denotes the mathematical expectation; is the integrated payoff function defined in step S4; apply momentum acceleration, calculate momentum update direction , and update the master car's strategy probability to , and denote the momentum accumulation variables of the master car and the competing car strategy probability distributions at the kth iteration, respectively; denotes the tensor product operation; S53. Calculate the norm of the change in the master vehicle strategy distribution between the current iteration and the previous iteration. If the norm of change is less than the convergence threshold or number of iterations If the iteration stops, output the current master vehicle strategy probability distribution. If the distribution is optimal, then let k = k + 1 and return to step S52 to continue iterating.

7. The intelligent driving lane-changing decision method for heavy-duty vehicles according to claim 6, characterized in that, Step S6 comprises: S61、After the iteration optimization process of step S5 ends, the output main vehicle strategy probability distribution Convergence verification is performed, and it is determined whether the output main vehicle strategy probability distribution satisfies and satisfies ; wherein, represents the final probability value of the main vehicle selecting the i-th action; N represents the total number of actions in the main vehicle action set; S62, selecting an optimal lane-changing action from the set of actions A of the host vehicle based on the probability distribution of the host vehicle strategy , by the principle of probability maximization : ； where argmax operation denotes selecting the action that maximizes the probability value as the optimal decision output;​ S63, determining the optimal lane change strategy of step S62 converts into a vehicle control instruction set , including a target lane number , a desired longitudinal acceleration , and a desired yaw rate , and outputs the generated vehicle control instruction to a drive-by-wire execution system of the vehicle, the vehicle control instruction being represented as: ； wherein, is a policy-instruction conversion function.

8. An intelligent driving lane-changing decision system for heavy-duty vehicles, characterized in that, The application comprises the following steps: The probabilistic game model construction module is configured to define the autonomous heavy vehicle as a host vehicle and identify a neighboring vehicle having a potential lane-changing conflict with the host vehicle as a competitor vehicle, construct a probabilistic game model involving the host vehicle, the competitor vehicle, and environmental disturbance, and define a state vector of the host vehicle, an action set of the host vehicle, and a state transition equation of the host vehicle in the probabilistic game model. The index definition module is configured to define a safety index based on a prediction result of the state transition equation, define an efficiency index based on a current speed of the host vehicle, and define a comfort index based on a current action characteristic in the action set of the host vehicle. The strategy entropy calculation module is configured to calculate a strategy probability distribution of each action in the action set of the host vehicle at a current time, calculate a strategy entropy based on the strategy probability distribution of the host vehicle, and calculate dynamic weights of the safety index, the efficiency index, and the comfort index based on the strategy entropy. The comprehensive payoff function construction module is configured to multiply the safety index, the efficiency index, and the comfort index by the corresponding dynamic weights, respectively, sum the results, and construct a comprehensive payoff function. The model iteration module is configured to use an alternating direction multiplier method layered optimization algorithm to iteratively update a strategy probability distribution of the host vehicle, a strategy probability distribution of the competitor vehicle, and a joint strategy distribution formed by the two strategy probability distributions, with the objective of minimizing the comprehensive payoff function, introduce a momentum acceleration mechanism in the iteration process to improve a convergence speed, and stop the iteration until a preset convergence condition is met. The lane-changing strategy output module is configured to select an action having the highest probability from the finally converged strategy probability distribution of the host vehicle as an optimal lane-changing strategy and output the optimal lane-changing strategy for controlling the host vehicle to perform a lane-changing operation.

9. A terminal, characterized by comprising: The program is executed by the processor to implement the method of any one of claims 1-7. The program is executed by the processor to implement the method of any one of claims 1-7. ​ ​ 10. A computer readable storage medium storing a computer program, characterized in that, ​

Images