A carbon emission reduction method for intelligent connected vehicles based on C-V2X vehicle-road collaboration

Abstract

The present invention belongs to the field of carbon emission reduction for intelligent connected vehicles, and particularly relates to a carbon emission reduction method for intelligent connected vehicles based on C‑V2X vehicle-road collaboration. The method includes the roadside unit (RSU) and the vehicle jointly acquiring road traffic information and vehicle information; the roadside unit (RSU) groups vehicles entering the non-lane-changeable area on a single lane before the intersection into a vehicle queue based on the road traffic information and the vehicle's information, and designates the vehicle closest to the intersection as the main vehicle (HV); the roadside unit (RSU) uses a decision-making model trained based on a deep reinforcement learning model to output the most economical driving action a for the vehicle based on road information and the main vehicle's information; the reward function used for training is designed to increase the vehicle's passing speed while avoiding frequent acceleration and deceleration and vehicle collisions; each vehicle implements carbon emission reduction driving based on this driving action. By optimizing vehicle driving behavior, traffic efficiency is improved and carbon dioxide emissions during vehicle driving are reduced.

Description

Technical Field

[0001] The present invention belongs to the field of carbon emission reduction of intelligent connected vehicles, and in particular relates to a carbon emission reduction method for intelligent connected vehicles based on C-V2X vehicle-road collaboration. Background Art

[0002] C-V2X (Cellular Vehicle-to-Everything) is a key enabling technology for smart transportation and intelligent driving. Through vehicle-to-vehicle and vehicle-to-road collaboration, it provides broader and more accurate information perception for these two areas at a lower cost, enabling more powerful connected intelligence. Currently, a growing number of studies are leveraging C-V2X technology to enhance driving safety, improve traffic efficiency, and support intelligent traffic management. C-V2X technology is also a solution for reducing road carbon emissions.

[0003] Due to the instability of urban road traffic flow, vehicles experience frequent acceleration and deceleration during operation. A method and system for coupling traffic operation and emission data based on VSP (Vehicle Specific Power) distribution (CN 104715605 A) analyzes traffic operation data and vehicle emission characteristics, improving the efficiency of road section emission assessments and demonstrating a correlation between vehicle driving behavior and vehicle emissions.

[0004] How to reduce road traffic carbon emissions while ensuring road traffic safety and driving efficiency is one of the key issues in the road traffic field. Summary of the Invention

[0005] In response to the problem of unreasonable acceleration, deceleration, and stop-and-go behavior of road vehicles, which increases road carbon emissions, the present invention designs a method for controlling intelligent connected vehicles to achieve carbon emission reduction based on C-V2X vehicle-road collaborative technology. Through vehicle-to-vehicle collaboration and vehicle-road collaboration, vehicle driving behavior is optimized to improve traffic efficiency and optimize road traffic flow to reduce carbon dioxide emissions during vehicle driving.

[0006] To address the above issues, the present invention provides a method for reducing carbon emissions from intelligent connected vehicles based on C-V2X vehicle-infrastructure collaboration, comprising the following steps:

[0007] Step 1: The roadside unit (RSU) and the vehicle jointly obtain road traffic information and vehicle information.

[0008] Step 2: The roadside unit (RSU) groups the vehicles that have entered the non-lane-changeable zone on a single lane before the intersection into a vehicle queue based on traffic information and vehicle information, and designates the vehicle closest to the intersection as the master vehicle (HV).

[0009] Step 3: The roadside unit (RSU) uses a decision-making model trained based on a deep reinforcement learning model to output the most economical driving action a for the vehicle based on road information and the vehicle's own information.

[0010] The state space A and action space S of the decision model are expressed as follows:

[0011] S={v,ac,v EV ,ac EV ,d,d TL ,s};

[0012] A = {v, ac};

[0013] Where v is the HV's own speed, ac is the acceleration of the main vehicle, and v EV is the speed of the vehicle in front of the HV, ac EV is the acceleration of the vehicle in front of the host vehicle, d is the distance between HV and the vehicle in front, d TL is the distance between the HV and the intersection, s is the current traffic light status, including the current state and the duration of the state;

[0014] The reward function used for training is designed to increase the vehicle's passing speed while avoiding frequent acceleration and deceleration and vehicle collisions;

[0015] Step 4: The roadside unit (RSU) sends the most economical driving maneuver to the host vehicle via I2V communication. The host vehicle then shares it with other vehicles in the platoon via V2V communication. Each vehicle then uses this driving maneuver as a basis for carbon emission reduction.

[0016] Step 5: After the current main vehicle passes the intersection, it is removed from the vehicle queue and the process returns to step 2 to continue.

[0017] Preferably, the reward function is expressed as follows:

[0018] R=r a +r b +r c

[0019]

[0020]

[0021]

[0022] Among them, r a is the vehicle low speed penalty function, r b is the acceleration penalty function, r c is the conditional reward function; α1 and α2 are the weights corresponding to the penalty function, T is the time step of each round in training, and vt is the current vehicle speed, v max Indicates the maximum speed without potential collision or the road speed limit, whichever is smaller, when the vehicle is traveling at high speed. t is the vehicle's acceleration at the current moment, ac max is the maximum acceleration of the vehicle.

[0023] Furthermore, the Actor-Critic algorithm framework is adopted in the training of the decision model.

[0024] Furthermore, during the training of the decision model, the proximal strategy optimization algorithm is used to update the model parameters.

[0025] Preferably, in step 4, if the vehicle is an autonomous driving vehicle, the most economical driving action of the vehicle is executed; if the vehicle is not an autonomous driving vehicle, the most economical driving action of the vehicle is prompted to the driver.

[0026] Furthermore, step 4 also includes calculating the carbon emission reduction of the vehicle after taking the most economical driving action based on the collected data in combination with the vehicle carbon emission model, and notifying the driver / passengers of the emission reduction.

[0027] Furthermore, the vehicle carbon emission model is as follows:

[0028]

[0029] β is the carbon emission coefficient, i.e. the amount of carbon dioxide emissions per unit energy consumption, f fuel,v is the unit distance fuel consumption factor in a certain speed range, v t is the driving speed at time t, γ is the compensation term for carbon dioxide emissions due to road congestion, and we have:

[0030]

[0031] t L is the total time the vehicle travels on this road section, t l For congestion time,

[0032] Due to the application of the above technical solutions, the present invention has the following advantages compared with the existing technology: the present invention adopts C-V2X technology to achieve a wider range of perception through vehicle-road collaboration, and enables vehicles to travel at an economical and stable speed through driving behavior suggestions, thereby optimizing road traffic flow and improving road traffic efficiency, reducing unreasonable acceleration and deceleration of vehicles, and effectively addressing the problem of increased carbon emissions caused by unreasonable driving behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 is a schematic diagram of an application scenario in an embodiment of the present invention;

[0034] Figure 2 is a communication message flow diagram in an embodiment of the present invention;

[0035] Figure 3 4 is a flow chart of a carbon emission reduction method in an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The present invention will be further described below with reference to the accompanying drawings and examples:

[0037] The application scenarios of this embodiment are as follows Figure 1 As shown in the figure, in urban road traffic scenarios, the road traffic efficiency is improved and road carbon dioxide emissions are reduced by controlling the vehicles on the road section between two intersections.

[0038] In this section of road, the road infrastructure is equipped with a complete road side unit (RSU), and all road vehicles are equipped with an on-board unit (OBU) and have V2X communication conditions. Figure 1 As shown in (a), the vehicle V1 represented by dark red is the main vehicle (HV) in V2I communication. The subsequent vehicles V2 and V3 obtain information from the main vehicle through V2V communication. The V2X data transmission in this example uses the PC5 interface (direct communication interface) of the cellular vehicle-to-everything (C-V2X) technology. PC5 has low latency and high reliability for short-distance, direct vehicle-to-vehicle communication, which enables vehicles to have a larger communication range and access to network resources. Figure 2 In the message flow framework shown, the OBU has its own fixed IP address and port number. The OBU constantly listens to port 30300 to collect broadcast messages and compares the messages sent by PC5 with the IP and port number of the host computer. If a match occurs, the OBU receives blind spot monitoring information from surrounding vehicles. If a match does not occur, the OBU discards the message and continues to listen for other messages on the port. In this example, the carbon emission reduction method for intelligent connected vehicles based on C-V2X vehicle-infrastructure collaboration achieves energy conservation and emission reduction by controlling the vehicle to exit the intersection smoothly and quickly, reducing unreasonable acceleration and deceleration.

[0039] This example provides a carbon emission reduction method for intelligent connected vehicles based on C-V2X vehicle-road collaboration. The process is as follows: Figure 3 As shown. It includes the following steps:

[0040] Step 1: Traffic data collection: sensors and other equipment obtain road information and vehicle surrounding information and encapsulate them into message sets;

[0041] Traffic data collection is completed by vehicle-mounted terminals with wireless communication capabilities, pedestrian portable devices and roadside sensing equipment. Each terminal can communicate through V2X to achieve comprehensive perception and collection of the vehicle itself and road traffic information.

[0042] Vehicle data collected by on-board sensors includes vehicle information: speed, acceleration, position, angular velocity, etc.

[0043] The vehicle's own data is collected by on-board sensors, and the vehicle's own condition and current location, surrounding vehicles, road environment and other information are perceived through sensors and technologies such as the Global Navigation Satellite System (GNSS). This places strict requirements on communication latency and reliability.

[0044] Roadside sensing equipment is primarily used to collect road traffic data, including multi-source sensors, traffic signal equipment, video surveillance equipment, and more. Information from each detection device must be compatible with the RSU communication interface. An open interface protocol must be set up on the RSU to collect and aggregate data, placing strict demands on communication latency and reliability.

[0045] The road traffic data aggregated by RSU includes: traffic light status, road congestion status, lane driving direction, etc.

[0046] In addition, RSU can also summarize road map data, including: road network information, intersection maps and road slopes, etc.

[0047] For infrastructure-to-vehicle (I2V) data transmission, the RSU obtains the surrounding intersection map from the map server and assembles it into a map message (MAP), including time, intersection information of the local area, road section information, lane information, and connection relationships between roads.

[0048] The RSU obtains the traffic light status and signal phase from the traffic light signal machine and assembles them into a traffic light message (SPAT) including the traffic light status, time and phase information.

[0049] RSU broadcasts MAP, SPAT and other messages through PC5 (direct communication interface).

[0050] The on-board unit OBU matches the message destination address through the IP and port number. If the IP and port number match, it receives the MAP and SPAT messages broadcast by the surrounding RSU through PC5, and parses them to obtain the road structure and traffic light information of the road ahead. If they do not match, it deletes the message and continues to listen to the port.

[0051] The vehicle matches the road information broadcasted in the RSU according to its own direction of travel, and associates the phase of the traffic light in front of the intersection and the distance from the vehicle to the intersection based on the collected information.

[0052] In vehicle-to-vehicle (V2V) data transmission, the vehicle's onboard unit (OBU) assembles information such as its own position, speed, gear, braking, heading angle, size, type, etc. into a basic safety message (BSM).

[0053] The OBU broadcasts the BSM message through PC5.

[0054] The BSM message is sent through the OBU transparent transmission module via fixed port number 30301. The message set is uploaded to the host computer computing unit for parsing the BSM message and obtaining information such as the location, status, and type of other vehicles. The vehicle obtains information about the road map, the traffic light ahead, the vehicle itself, and surrounding vehicles.

[0055] Step 2: The roadside unit (RSU) organizes the vehicles entering the no-lane-change zone on a single lane before the intersection into a vehicle platoon based on road traffic information and vehicle information. The vehicle closest to the intersection is designated as the lead vehicle (HV). Because vehicle position information and road information (including lane divisions, specific no-lane-change zone divisions, intersection location, etc.) are readily available, the RSU can easily form the vehicle platoon. The specific method is not detailed here.

[0056] Step 3: The roadside unit (RSU) uses a decision-making model trained based on a deep reinforcement learning model to output the most economical driving action a for the vehicle based on road information and the vehicle's own information.

[0057] To obtain the final decision-making model based on the deep reinforcement learning model, this example first initializes the road length, traffic light locations, vehicle directions, number of vehicles on the road, and basic driving rules in the simulation environment. Then, an agent corresponding to the main vehicle (HV) and the environment (Environment) in which it interacts are established. The agent's action space and state space are designed based on road traffic light information and vehicle driving scenarios.

[0058] The vehicle action space A includes the vehicle's own velocity v and acceleration ac. As you can imagine, the driving action of the HV is continuous, so the state space in this example is a discrete state space. For example, the driving action a = {v = 5m / s, ac = 3m / s 2};

[0059] The state space S includes its own velocity v, acceleration ac, and the velocity v of the vehicle in front of HV EV and acceleration ac EV , the distance d from the vehicle in front, the distance d from the vehicle to the nearest intersection TL The current traffic light state sp includes three states: red, green, and yellow, as well as the duration of each state. For example, the current green light and the duration of the green light are:

[0060] s={v,ac,v EV ,ac EV ,d,d TL ,sp}.

[0061] The deep reinforcement learning model in this example uses the Actor-Critic algorithm framework, and the Actor network implements the policy function π with parameters θ θ Used to select action a, the Critic network is used to calculate the value function V of the current state π (s). In addition, the reward function R of the intelligent agent is designed according to the actual carbon emission reduction target requirements and vehicle driving scenarios as follows:

[0062] R=r a +r b +r c

[0063]

[0064]

[0065]

[0066] Among them, r a is the vehicle low speed penalty function, r b is the acceleration penalty function, r c is the conditional reward function; each action reward is the sum of three types of penalties / rewards, α1 and α2 are the weights corresponding to the penalty function, which can be set during training to obtain the optimal speed, and T is the time step of each round. t is the current vehicle speed, v max Indicates the maximum speed without potential collision or the road speed limit, whichever is smaller, when the vehicle is traveling at high speed. t is the vehicle's acceleration at the current moment, ac max is the maximum acceleration of the vehicle. a Allowing vehicles to travel at their maximum speed can improve road traffic efficiency; b The Euclidean norm after normalization of the acceleration ac is used to keep the vehicle's acceleration stable, avoid frequent acceleration and deceleration, and improve fuel economy; c This is to ensure driving safety, avoid collisions that cause road congestion, and prevent vehicle idling from increasing road carbon emissions.

[0067] Please note that the RSUs and vehicles required for model training and establishment can be obtained. The specific acquisition method has been introduced in step 1 and will not be repeated here.

[0068] In this example, the policy gradient and importance sampling methods of the Proximal Policy Optimization Algorithms (PPO, DOI: 10.48550 / arXiv.1707.06347) in the existing literature are used to update the parameter θ, and the intelligent agent is trained by optimizing the strategy to maximize the long-term return.

[0069] In the specific training implementation, the algorithm parallel rounds are set to E = 16, the number of time steps in each round is T = 256, the training termination iteration number I = 1024, and the minimum iteration number K = 32.

[0070] First, in each round of operation, starting from the first time step, the input state s1, the Actor network selects action a1, the Critic calculates the state value, the agent performs the action while interacting with the environment, the environment feedback rewards and enters the new state s2, the Actor network selects an action again based on the new state, and so on. In each state, the agent will make an action, and after one round, a state-action sequence (Trajectory) is formed:

[0071] τ={s1,a1,s2,a2,....,s t ,a t ,....,s T ,a T},

[0072] Under the policy parameter θ, the probability of each round sequence is expressed as P θ (τ) indicates that the probability of a sequence τ occurring can be expressed as:

[0073] p θ (τ)=p(s1)p(a1|s1)p(s2|s2,a1)p(a2|s2)p(s3|s2,a2)...

[0074] Therefore, the goal of strategy optimization is to maximize the following expected reward:

[0075] in,

[0076] When implementing the PPO algorithm in this example, the clipping (Clipped Surrogate Objective) method proposed in the literature is used to limit the policy update amplitude. The hyperparameter ε used for clipping is initially set to 0.2 (which can be adjusted according to the training status). This limits the policy update amplitude in each round and also limits some data with large deviations, making the model converge faster.

[0077] In specific training, the GAE (Generalized Advantage Estimation) method is used to estimate the advantage of each action relative to the average action (advantage function) At the same time, the error obtained by subtracting the estimated value of the Critic network from the actual reward is used as the loss function of the Critic network to update the parameters of the Critic network. This error is related to Together, they are used to calculate the policy gradient and update the parameters of the Actor network based on the importance sampling method.

[0078] According to the method in the literature, data collection, advantage estimation and strategy update are iterated until the number of iterations I is reached or the model training converges. According to the above settings, 16 rounds of training are carried out in parallel based on the current network parameters each time, and 16 state-action sequences are obtained. At this time, the cumulative reward is selected. The largest sequence is used to update the network parameters. Since the minimum number of iterations is set, the network parameters are updated at least K=32 times.

[0079] Finally, the agent is trained to obtain a stable Actor network model for the speed control decision model. The trained decision model is used to calculate the economic speed for the moving vehicle, so that the vehicle can recommend the most economical and smooth vehicle speed and acceleration under different driving states s.

[0080] Step 4: The roadside unit (RSU) sends the most economical driving maneuver to the host vehicle via I2V communication. The host vehicle then shares it with other vehicles in the platoon via V2V communication. Each vehicle then implements carbon emission reduction driving based on this driving maneuver.

[0081] If the main vehicle / other vehicle is an autonomous vehicle, the autonomous driving decision maker will be informed of the most economical and smooth vehicle speed and acceleration, and the vehicle will then drive directly at the most economical speed and acceleration. If the main vehicle / other vehicle is an assisted driving vehicle, the speed recommendation will be displayed on the on-board display screen to remind the driver of the recommended speed of the road ahead. If acceleration measurement is required at this time, the driver can also be prompted in graphical form whether the current vehicle acceleration has reached or exceeded the most economical acceleration when accelerating, and the gap between the current acceleration and the most economical acceleration can be visualized.

[0082] Step 4: The roadside unit (RSU) sends the most economical driving maneuver to the host vehicle via I2V communication. The host vehicle then shares it with other vehicles in the platoon via V2V communication. Each vehicle then uses this driving maneuver as a basis for carbon emission reduction.

[0083] Step 5: After the current main vehicle passes the intersection, it is removed from the vehicle queue and the process returns to step 2 to continue.

[0084] like Figure 1 As shown in (b), after the current vehicle V1 leaves the intersection, the vehicle agent is switched to control the vehicle agent closest to the intersection in the new state, while the following vehicles continue to follow. This effectively reduces road vehicle resources and model training time by only controlling the vehicle closest to the intersection.

[0085] The following simulation experiment calculation compares the carbon emission reduction of roads with and without emission reduction measures.

[0086] In a 5KM road section scenario, the proportions of small passenger cars, buses, medium-sized buses, medium-sized trucks, and large trucks are set to 50%, 20%, 10%, 15%, and 5% respectively. Among them, 10% of the buses are set at stations on the road section and need to stop, and the other 10% are in normal driving. The speed of large trucks is weighted according to the load. The simulation experiment is carried out under the condition of traffic flow Q=4096. The vehicles in Group 1 use the economic speed model of the present invention, and the vehicles in Group 2 do not perform any control and drive as ordinary vehicles without driving optimization. The driving time of Group 1 and Group 2 is counted, and the carbon emissions of Group 1 and Group 2 are calculated.

[0087] The carbon emissions corresponding to the speed of a bicycle on this road are calculated as follows: β is the carbon emission coefficient, i.e. the amount of carbon dioxide emissions per unit energy consumption, f fuel,v is the unit distance fuel consumption factor in a certain speed range (this parameter can be found in the standard document), v t is the driving speed at time t. γ is the compensation term for carbon dioxide emissions due to road congestion, which is:

[0088]

[0089] t L is the total time the vehicle travels on this road section, t l is the congestion time, E = β·f·L, f is the fuel consumption per unit distance, and L is the length of the current road section.

[0090] By adding up the carbon dioxide emissions of all vehicles on the road (i.e., traffic flow), we can get the carbon emissions under a certain traffic flow. The calculation formula is: The carbon emissions of Group 2 were compared with those of Group 1. The experimental results showed that carbon dioxide emissions of different types of vehicles were reduced, with a reduction of 19% for small passenger cars and 13.90% for buses. This may be related to the behavior of buses stopping at stations before joining the fleet. The average carbon dioxide emissions on the road were reduced by 16%.

[0091] It is worth noting that in some embodiments, the carbon emissions calculated using the above-mentioned emission model, and the carbon emission reduction after the current vehicle adopts the most economical driving action, which can be further calculated, are calculated and displayed to the driver / passengers of the vehicle, so that the driver / passengers can feel that they have participated in energy conservation and emission reduction and gain a sense of accomplishment, while encouraging drivers of non-autonomous driving vehicles to drive according to the suggested most economical driving actions.

[0092] The present invention aims to address the problem of unreasonable driving behavior of road vehicles, which causes increased carbon emissions on traffic congested roads, and provide a solution for reducing carbon dioxide emissions during vehicle driving.

Claims

1. A carbon emission reduction method for intelligent connected vehicles based on C-V2X vehicle-road collaboration, characterized in that: The following steps are involved: Step 1: The roadside unit (RSU) and the vehicle jointly obtain road traffic information and vehicle information. Step 2: The roadside unit (RSU) groups the vehicles that have entered the non-lane-changeable zone on a single lane before the intersection into a vehicle queue based on traffic information and vehicle information, and designates the vehicle closest to the intersection as the master vehicle (HV). Step 3: The roadside unit (RSU) uses a decision-making model trained based on a deep reinforcement learning model to output the most economical driving action a for the vehicle based on road information and the vehicle's own information. The state space A and action space S of the decision model are expressed as follows: ， ， in, The main vehicle's own speed, is the acceleration of the main vehicle, is the speed of the vehicle in front of the host vehicle, is the acceleration of the vehicle in front of the host vehicle, is the distance between the host vehicle and the vehicle in front, is the distance between the main vehicle and the intersection, The current status of the signal light, including the current status and the duration of the status; The reward function used for training is designed to increase the vehicle's passing speed while avoiding frequent acceleration and deceleration and vehicle collisions. The reward function is expressed as follows: ， ， ， ， in, is the vehicle low speed penalty function, is the acceleration penalty function, is the conditional reward function; is the weight corresponding to the penalty function, is the time step of each round in training, is the vehicle speed at time t, Indicates the maximum speed without potential collision in a high-speed driving scenario or the road speed limit, whichever is smaller; is the acceleration of the vehicle at time t, is the maximum acceleration of the vehicle; Step 4: The roadside unit (RSU) sends the vehicle's most economical driving maneuver to the host vehicle via I2V communication. The host vehicle then shares it with other vehicles in the platoon via V2V communication. Each vehicle then uses this most economical driving maneuver as a basis for carbon emission reduction. Step 5: After the current main vehicle passes the intersection, remove it from the vehicle queue and return to step 2 to continue execution.

2. The method according to claim 1, characterized in that The Actor-Critic algorithm framework is used in the training of the decision model.

3. The method according to claim 2, characterized in that During the training of the decision model, the proximal strategy optimization algorithm is used to update the model parameters.

4. The method according to claim 1, wherein In step 4, if the vehicle is an autonomous driving vehicle, the most economical driving action of the vehicle is executed; if it is not an autonomous driving vehicle, the most economical driving action of the vehicle is prompted to the driver.

5. The method according to claim 1, wherein Step 4 also includes calculating the vehicle's carbon emission reduction after taking the most economical driving action based on the collected data and combining it with the vehicle carbon emission model, and notifying the driver / passengers of the emission reduction.

6. The method according to claim 5, characterized in that The vehicle carbon emission model is as follows: ， is the carbon emission coefficient, i.e. the amount of carbon dioxide emissions per unit energy consumption. is the unit distance fuel consumption factor in a certain speed range, is the driving speed at time t, Compensation items for carbon dioxide emissions from road congestion include: ， is the total time the vehicle travels on this road section, For congestion time, .

Images