Self-adaptive parking lot changeable entrance and exit control method

Abstract

A self-adaptive parking lot changeable entrance and exit control method, comprising: receiving historical vehicle data at entrance and exit barrier gates of a parking lot (step a); on the basis of the historical vehicle data at the entrance and exit barrier gates of the parking lot, predicting the number of vehicles leaving the parking lot in a unit time and the number of vehicles entering the parking lot in the unit time (step b); calculating the minimum value of the number of exit lanes lot and the minimum value of the number of entrance lanes required by the parking (step c); using queuing information of an upstream road section at the entrance and exit of the parking lot, queuing information of a downstream road section of the entrance and exit of the parking lot, queuing information of a road section inside the parking lot, the number of vehicles leaving the parking lot in the unit time, and the number of vehicles entering the parking lot in the unit time as input, and using the minimum value of the number of exit lanes and the minimum value of the number of the entrance lanes of the parking lot, and an entrance and exit lane setting principle as constraint conditions, constructing a parking lot entrance and exit lane setting model, and training the model using interaction data (step d); in the trained model, inputting the real-time information of the queuing information of the downstream road section at the entrance and exit of the parking lot, the queuing information of the road section in the parking lot, the number of vehicles leaving the parking lot in the unit time, and the number of vehicles entering the parking lot in the unit time, and by using the entrance and exit lane setting model of the parking lot, outputting the setting of each entrance and exit lane of the parking lot (step e).

Description

An adaptive parking lot variable entrance and exit control method Technical Field

[0001] The present invention relates to the technical field of parking facility control, and in particular to an adaptive parking lot variable entrance and exit control method based on traffic state perception and intelligent management and control. Background Art

[0002] With the development of the automotive industry, the increasing demand for motor vehicle travel has not only put pressure on road traffic but also created difficulties in parking system management and control. Large urban parking lots often have multiple entrances and exits, all of which follow the same configuration method. This leads to congestion at some entrances and exits while leaving others idle. The proper configuration of entrances and exits has become a major challenge in managing and controlling traffic during peak hours. The number of parking lot entrances and exits determines whether traffic demand can be met, while their distribution determines whether vehicles can enter and exit the parking lot efficiently. Improper entrance and exit configuration not only reduces user experience but also places significant pressure on the dynamic traffic flow outside the parking lot. While existing parking lot entrance and exit configuration methods can integrate information such as traffic volume, parking lot size, and the grade of the road connecting the entrances and exits, traffic systems are dynamic and often exhibit significant tidal phenomena. If parking lot entrance and exit configurations are based solely on limited, fixed traffic information, it will be difficult to meet the needs of vehicles entering and leaving the lot, resulting in low traffic efficiency and even traffic paralysis. At the same time, parking lot entrances and exits serve as connecting points between dynamic and static traffic. The previous fixed parking lot entrance and exit setting methods lack adaptability. After the traffic system is impacted, it is difficult to quickly return to normal levels, and the space for flexible adjustment of the system is extremely limited.

[0003] At present, the existing parking lot entrance and exit setting methods are mainly:

[0004] 1. Parking lot entrance and exit setting method based on the number of parking spaces

[0005] The "Standard for Traffic Design and Parking Garage (Lot) Establishment of Construction Engineering" (DGJ08-7-2006) stipulates that when the number of parking vehicles is less than 100, the entrance and exit of the motor vehicle parking garage shall be set up with no less than one double lane or two single lane exits; when the number of parking vehicles is greater than or equal to 100 and less than 200, no less than two single lane exits shall be set up; when the number of parking vehicles is greater than or equal to 200 and less than 700, no less than two lanes for entry and two lanes for exit shall be set up; when the number of parking vehicles is greater than or equal to 700, no less than three double lane exits shall be set up, and a service level evaluation shall be conducted to determine the number of entrances and exits.

[0006] 2. Parking Lot Entrance and Exit Setting Method Based on Queuing Theory

[0007] Assuming that the traffic flow in and out of a parking lot follows a Poisson distribution, the time it takes for a vehicle to pass through the barrier follows a negative exponential distribution, and the total number of barrier gates is equivalent to the number of service stations, the time a vehicle spends in a parking space can be considered infinite. Therefore, the process of entering and exiting the parking lot can be modeled using queuing theory. For a plot with multiple entrances and exits, the queuing system can be viewed as a system consisting of multiple M / M / 1 or M / M / S systems. In practical applications, it is necessary to combine peak traffic volume and the time spent to calculate the peak hour arrival rate of vehicles and the average service time of parking lot exit and entrance barriers. This allows the calculation of the minimum number of exit and entrance lanes. Queuing theory is then used to verify whether the minimum number of lanes meets the service requirements, i.e., a pre-set threshold. If not, an additional lane is added until the verification passes.

[0008] 3. Considering the connection between the road and the parking lot entrance and exit

[0009] When a parking lot entrance and exit connects to a one-way road, the entrance and exit roads should be designed to avoid intersecting traffic. Typically, the entrance road is placed first, followed by the exit road, in the direction of the road, to minimize the impact of vehicles entering and exiting the parking lot on the road. When a parking lot entrance and exit connects to a two-way road, the principle is to ensure that traffic entering and exiting the parking lot does not intersect with traffic on the road, and that vehicles turn right into the parking lot. When a parking lot entrance and exit are located on one side of an intersection, the entrance and exit roads are typically arranged in a clockwise direction, especially if the entrance and exit are on different roads.

[0010] The aforementioned methods for setting up parking lot entrances and exits are all based on rules that set the number and location of parking lot entrances and exits. The main process is to calculate the minimum number of parking lot exit and entrance lanes based on the number of parking spaces within the parking lot, the peak hour arrival rate of vehicles, the average service time of the parking lot exit and entrance gates, and the connected roads (Figure 1). Combined with queuing theory verification (Figure 2), and based on meeting the minimum number of parking lot exit and entrance lanes, appropriately increase lanes to meet service level requirements. When setting the location of exit and entrance lanes, it is necessary to consider the relationship between the parking lot entrance and exit and the connecting roads and set them based on meeting the entrance and exit lane setting principles (Figure 3).

[0011] Existing technology

[0012] CN114758497A

[0013] Explanation of related terms

[0014] 1. Variable entrance and exit control of parking lot: The entrance and exit lanes of the parking lot adopt a variable lane setting mode. The lane can be changed to an exit lane or an entrance lane. The control is completed according to the entrance and exit setting plan output by the parking lot entrance and exit lane setting model.

[0015] 2. Dynamic traffic information: Dynamically changing traffic information outside the parking lot, specifically referring to the actual queue information on the upstream section of the parking lot entrance and exit, the queue information on the downstream section of the parking lot entrance and exit, the number of vehicles leaving the parking lot per unit time, and the number of vehicles entering the parking lot per unit time.

[0016] 3. Static traffic information: Traffic information within the parking lot, specifically the queue information of the road sections within the parking lot, which can be expressed by the average queue length of the road section and the average speed of vehicles on the road section.

[0017] 4. Parking lot external road sections: upstream sections of the parking lot entrance and exit, and downstream sections of the parking lot entrance and exit.

[0018] 5. Upstream section of the parking lot entrance and exit: The road section between the parking lot entrance and exit and the upstream intersection is bounded by the two intersections where the road where the parking lot entrance and exit intersects with other roads.

[0019] 6. Downstream section of parking lot entrance and exit: The road section between the parking lot entrance and exit and the downstream intersection is bounded by the two intersections where the road where the parking lot entrance and exit intersects with other roads.

[0020] 7. Parking lot internal road sections: road sections where vehicles travel inside the parking lot, and road sections where vehicles queue and / or travel at parking lot entrances and exits.

[0021] 8. Vehicle data inside and outside the parking lot: positioning coordinates, speed information, detection time, license plate number, appearance characteristics (including model, color, etc.), and lane.

[0022] 9. Vehicle data at parking lot entrance and exit gates: license plate number, appearance characteristics (including vehicle model, color, etc.), inspection time (including specific day of the week and time period), weather, and parking fee.

[0023] 10. Queue information: average queue length on the road section and average vehicle speed on the road section.

[0024] 11. Queue length: the number of vehicles traveling at a speed lower than 0.1 m / s.

[0025] 12. Number of vehicles leaving the parking lot per unit time and number of vehicles entering the parking lot per unit time: the number of vehicles leaving and entering the parking lot per unit time.

[0026] 13. Parking lot entrances and exits: The locations of parking lot exits and entrances can be single lanes, double lanes, or multiple lanes. The available settings for each lane are exit lanes, entrance lanes, or closed lanes.

[0027] 14. Parking lot entrance and exit lane settings: The number and location of parking lot exit and entrance lanes. The location indicates that the exit and entrance lanes are specifically set on a certain lane of a certain entrance and exit.

[0028] 15. Minimum number of parking lot exit lanes and entrance lanes required: The minimum number of parking lot exit lanes and entrance lanes required to meet the basic needs of vehicles leaving and entering the entire parking lot.

[0029] 16. Parking Lot Entrance and Exit Lane Configuration Model: This model takes queue information for upstream and downstream sections of parking lot entrances and exits, queue information for internal sections of parking lot entrances and exits, the number of vehicles leaving the parking lot per unit time, and the number of vehicles entering the parking lot per unit time as inputs. It also uses the minimum number of required parking lot exit lanes and entrance lanes, as well as the parking lot entrance and exit lane configuration principles, as constraints to output a model for parking lot entrance and exit lane configuration.

[0030] 17. Interactive data: In the parking and road traffic simulation platform, the parking lot entrance and exit model is set according to the traffic status s of the parking lot entrance and exit i observed by it. i , output the entrance and exit setting plan a of the parking lot entrance and exit i i , after being executed in the simulation environment, the traffic state of the parking lot entrance i changes to state s i ', the traffic status data before and after s i and s i '、Parking lot entrance and exit setting plan a i , the evaluation results r of the parking lot entrance and exit setting scheme are recorded, and then the overall state of the parking lot entrance and exit is set to x=[s1,...,s N ], the overall parking lot entrance and exit control scheme is set as a=[a1,...,a N ], defined as interaction data<x,a,r,x'> , where x'=[s1',...,s N '].

[0031] 18. Experience pool: A database that stores interaction data and can provide interaction data collected at different times for model training.

[0032] 19. Reward function for the parking lot entrance and exit lane configuration model: Based on the overall parking lot entrance and exit states x and x' observed by the platform, and the currently used parking lot entrance and exit lane configuration plan a, the output of the parking lot entrance and exit lane configuration model is a function f(x, a, x') that measures the quality of the parking lot entrance and exit lane configuration plan. r is the reward, i.e., the calculation result of the reward function.

[0033] 20. Historical data: Records of vehicles entering and exiting the parking lot at all entrances and exits, including time, entrance and exit location, vehicle type, weather, and parking fee information. The data of vehicles entering and exiting are recorded separately.

[0034] 21. Machine Learning Algorithms: These include machine learning algorithms that can combine big data to achieve predictions, such as random forests, support vector machines, and neural networks.

[0035] 22. Four-stage forecasting: A method used in the transportation field to forecast travel demand, including four stages: traffic generation, traffic distribution, traffic mode classification, and traffic volume allocation.

[0036] 23. Land use similarity: The land use type above the underground parking lot is determined according to the "Urban Land Classification and Planning and Construction Land Standards". The area occupied by each land use type is determined. The area occupied by the land use type k of the current parking lot is The land use area of ​​a certain type k of built parking lot is The ratio of the two Defined as the land use similarity, if all land use types of the current parking lot and the reference parking lot satisfy δ min ≤δ≤δ max If the threshold requirement is met, the current parking lot and the reference parking lot are considered to be similar, and the historical data of the reference parking lot can be used to predict the number of vehicles leaving the current parking lot per unit time and the number of vehicles entering the parking lot per unit time.

[0037] 24. Empirical coefficient: The empirical coefficient can be the similarity of a particular land use type or the average of the similarities of all land use types, with reference to completed parking lots whose land use similarity meets the threshold requirements.

[0038] 25. Multi-Agent Deep Deterministic Policy Gradient Network: An algorithm used in deep reinforcement learning algorithms to solve collaborative decision-making problems among multiple agents.

[0039] 26. Critic Network: A network in a multi-agent deep deterministic policy gradient network that takes the observed state and parking lot entrance and exit lane configuration options as input and outputs the expected benefit. It is used to evaluate the pros and cons of the selected parking lot entrance and exit lane configuration options under the current state.

[0040] 27. Performer Network: A network in a multi-agent deep deterministic policy gradient network that takes the observed state as input and outputs parking lot entrance and exit lane configuration options. It is used to select parking lot entrance and exit lane configuration options.

[0041] 28. Deep Deterministic Policy Gradient Network: An algorithm used in deep reinforcement learning algorithms to solve decision-making problems for single agents. By establishing two critic and performer networks to directly output actions and their quality evaluations, it can effectively solve optimization problems with continuous action spaces.

[0042] 29. Deep Q-Network: An algorithm used in deep reinforcement learning algorithms to solve the decision-making of a single intelligent agent. It relies solely on neural networks to complete action evaluation and find the optimal action.

[0043] 30. Convergence: When the parking lot entrance and exit lane configuration model reaches a late stage of training, a reward function is used to evaluate the pros and cons of the parking lot entrance and exit lane configuration scheme. A high reward and a difference in rewards between adjacent training rounds within the required numerical range indicate that the model has been trained to convergence.

[0044] 31. Electronic display screens in parking lots: Wall-mounted or suspended electronic display screens are installed on the parking lot structural columns or at the top of the driveway to dynamically display changes in the parking lot entrance and exit settings, timely adjust the vehicle guidance strategy inside the parking lot, and change the parking lot exit direction guidance.

[0045] 32. Roadside variable information board: An electronic display screen is installed on the roadside outside the parking lot. Combined with graded guidance, it informs passing vehicles within a certain range around the parking lot of the parking lot entrance and exit settings and the number of vacant parking spaces.

[0046] 33. Navigation Application: For vehicles that need to exit or enter a parking lot, the mobile or vehicle-based navigation application will autonomously select the best parking lot entrance or exit, based on the vehicle's location and destination, and complete the corresponding optimal route planning to guide the driver in controlling the vehicle.

[0047] Summary of the Invention

[0048] The purpose of the present invention is to provide an adaptive parking lot variable entrance and exit control method, which comprehensively considers dynamic traffic information and static traffic information, dynamically adjusts the parking lot entrance and exit setting plan, and guides vehicles to change their driving paths to reach their destinations via recommended parking lot entrances and exits, thereby improving the efficiency of vehicles leaving and entering the parking lot, reducing queues inside and outside the parking lot, avoiding traffic congestion caused by uneven spatial distribution of traffic flow, and comprehensively improving the operating efficiency and service level of dynamic and static traffic systems.

[0049] An adaptive parking lot variable entrance and exit control method changes the number and position of the exit and entrance lanes of the parking lot entrance and exit by dynamically adjusting the parking lot entrance and exit lane setting scheme, and guides vehicles to change their driving paths in combination with traffic induction measures. The main process is shown in Figure 4, and the main steps include: receiving historical vehicle data at the parking lot entrance and exit gates, predicting the number of vehicles leaving the parking lot per unit time and the number of vehicles entering the parking lot per unit time, calculating the minimum number of required parking lot exit lanes and the minimum number of entrance lanes, training the parking lot entrance and exit lane setting model, and applying the parking lot entrance and exit lane setting model. Optionally, when applying the parking lot entrance and exit lane setting model, combined with internal and / or external induction of the parking lot, guide the driver to drive along the re-planned path. The solution of the present invention includes the following steps:

[0050] a) Receive historical vehicle data at the parking lot entrance and exit gates

[0051] At the gate of the parking lot entrance i, the camera and vehicle target recognition and feature extraction technology are used to obtain the information of vehicle v, including the license plate number Appearance features (including model, color, etc.) Detection time And record the detection time (Include specific day of the week and time period ) and weather information Calculate parking costs Then all the vehicle data collected at the gate is recorded as D p , where the historical vehicle data of vehicle information collected by the camera at the entrance and exit of the parking lot is represented by D p,his The real-time vehicle data collected by the camera at the entrance and exit of the parking lot is represented by D p,real .

[0052] b) Prediction of the number of vehicles leaving the parking lot per unit time and the number of vehicles entering the parking lot per unit time

[0053] Using the historical data collected at each gate at the entrance and exit of the parking lot p,his , predict the number of vehicles leaving the parking lot per unit time V out,fu , the number of vehicles entering the parking lot per unit time V in,fu If the parking lot has more than one year of historical data D p,his , the number of vehicles leaving the parking lot per unit time V out,fu , the number of vehicles entering the parking lot per unit time V in,fu Determined by machine learning algorithm; if the historical data D of the parking lot p,his In less than one year, the number of vehicles leaving the parking lot per unit time V out,fu, the number of vehicles entering the parking lot per unit time V in,fu Based on the four-stage prediction method, the reference land use property similarity δ is between δ l to δ a Historical data D of built parking lots f f,his The number of vehicles leaving the parking lot per unit time and the number of vehicles entering the parking lot per unit time of parking lot f are multiplied by the empirical coefficient λ to determine the number of vehicles leaving the parking lot per unit time V of parking lot p. out,fu and the number of vehicles entering the parking lot per unit time V in,fu The empirical coefficient λ can be the similarity of a certain land use type or the mean of the similarities of all land use types.

[0054] c) Calculation of the minimum number of required parking lot exit lanes and entrance lanes

[0055] According to the predicted number of vehicles leaving the parking lot per unit time V out,fu , the number of vehicles entering the parking lot per unit time V in,fu Calculate the minimum number of parking lot exit lanes c required out and the minimum number of entrance channels c in , ensuring that when the parking lot entrance and exit settings change, at least the needs of vehicles leaving and entering the lot are met.

[0056] d) Parking lot entrance and exit setting model training

[0057] In the traffic simulation software, roads and parking spaces are established based on the actual road traffic network, parking lot entrance and exit locations, internal sections of parking spaces, and parking space layout. The driver's starting and ending points and route selection are simulated based on the predicted traffic demand. The interaction logic between the simulation software and the background control program is established, and the background control program can then obtain real-time queue information on the upstream section of the parking lot entrance and exit. up,sim and downstream section queue information q down,sim , parking lot internal road section queue information park,sim , the number of vehicles leaving the parking lot per unit time V out,sim , the number of vehicles entering the parking lot per unit time V in,sim , and optimize the parking lot entrance and exit setting plan in real time based on the parking lot entrance and exit setting model.

[0058] In traffic simulation, the queue information of the upstream road section of the parking lot entrance and exit transmitted back by the simulation platform in real time is used. up,sim , Queue information of the road section downstream of the parking lot entrance and exit q down,sim , parking lot internal road section queue information park,sim , the number of vehicles leaving the parking lot per unit time V out,sim , the number of vehicles entering the parking lot per unit time V in,simAs state x, the minimum number of parking lot exit lanes c required out and the minimum number of entrance channels c in , parking lot entrance and exit setting principles as constraints, based on deep reinforcement learning algorithms (including deep Q networks, deep deterministic policy gradient networks, multi-agent deep deterministic policy gradient networks, etc.), with the minimum number of required parking lot exit lanes and the minimum number of entrance lanes, and parking lot entrance and exit setting principles as action constraints, a parking lot entrance and exit setting model M0 is constructed, and a parking lot entrance and exit setting plan a is output. The plan is then implemented at the parking lot entrance and exit in the simulation platform, and the state x' of the parking lot entrance and exit at the next moment is obtained. The reward r is used to evaluate the plan a, and the interaction data is used to calculate the reward r.<x,a,r,x'> The parking lot entrance and exit road setting model is trained until convergence, and a trained parking lot entrance and exit road setting model M1 is obtained.

[0059] The reward r of the parking lot entrance and exit lane setting model is expressed by the average queue length of the road section within a certain period of time, the driving efficiency, and whether the parking lot entrance and exit lane setting principles are met. For the case where the constraints are not met, after obtaining the parking lot entrance and exit lane setting plan, if the entrance and exit lane setting plan meets the entrance and exit lane setting principles, the number of exit lanes and entrance lanes of all parking lot entrances and exits is counted, and the statistical number is obtained. If the statistical number is lower than the minimum number of exit lanes c required by the parking lot, the reward r is expressed by the average queue length of the road section within a certain period of time, the driving efficiency, and whether the parking lot entrance and exit lane setting principles are met. out and the minimum number of entrance channels c in , then the exits and / or entrances are randomly opened on the basis of meeting the entrance and exit setting principles, and the number of exits and entrances meets the minimum requirements of each; otherwise, the entrances and exits are arranged according to the entrance and exit setting principles.

[0060] e) Application of parking lot entrance and exit model

[0061] On road section R, millimeter-wave radar is used to obtain vehicle data within the radar field of view. Can include world geographic coordinates Speed ​​information Detection time At the same time, the camera is used to obtain vehicle video image data within the camera field of view Including the vehicle license plate number obtained based on vehicle target recognition and feature extraction technology Appearance features (including model, color, etc.) Positioning coordinates Detection time Lane Based on the positioning coordinates of the millimeter wave radar and the camera, the positioning coordinates of the vehicle are converted into a specific lane and a specific position on the lane, and then the information of the vehicle v (including the license plate number) is converted into a specific lane and a specific position on the lane. Appearance characteristics Lane coordinate speed ) and detection time Integrate and record together, and further extract the queue information of the upstream section of the parking lot entrance and exit according to the vehicle speed, lane and detection time up,real , Queue information of the road section downstream of the parking lot entrance and exit q down,real , parking lot internal road section queue information park,real At the same time, the real-time detection data D collected by the camera at the entrance and exit of the parking lot p,real In the example, based on the number of vehicles entering and exiting the parking lot in the past hour, the real-time number of vehicles leaving the parking lot per unit time V is obtained. out,real , the number of vehicles entering the parking lot per unit time V in,real .

[0062] After the parking lot entrance and exit model training is completed, the real-time parking lot entrance and exit upstream section queue information q is input into the trained model M1 up,real , Queue information of the road section downstream of the parking lot entrance and exit q down,real , parking lot internal road section queue information park,real , the number of vehicles leaving the parking lot per unit time V out,real , the number of vehicles entering the parking lot per unit time V in,real , output parking lot entrance and exit setting plan a real .

[0063] f) Parking lot internal and / or external guidance (optional)

[0064] After the parking lot entrance and exit lane configuration plan changes, if the vehicle is inside the parking lot, on the one hand, the updated parking lot exit lane configuration is obtained through the vehicle-to-road communication equipment inside the parking lot and loaded into the high-precision map application, realizing the re-planning of the driving path of the vehicle-side navigation application; on the other hand, the electronic display screen in the parking lot provides real-time prompts of the vehicle's driving direction, realizing real-time navigation within the parking lot and helping the driver find the parking lot exit. After the parking lot entrance and exit lane configuration plan changes, if the vehicle needs to enter the parking lot from outside the parking lot, the driver can view the updated parking lot entrance and exit configuration plan and the number of available parking spaces on the variable information board on the roadside, or use communication technology to communicate with the roadside smart facilities to obtain this information and load it into the high-precision map application, realizing the re-planning of the driving path of the vehicle-side navigation application. The vehicle will go to the corresponding parking lot entrance according to the navigation information on the map application.

[0065] Compared with the prior art, the present invention has the following beneficial effects:

[0066] (1) The parking lot entrance and exit road settings of the present invention comprehensively consider the congestion conditions of the upstream and downstream sections of the parking lot entrance and exit and the congestion conditions of the sections inside the parking lot, control the parking lot entrance and exit road settings in real time, reasonably distribute the inflow and outflow traffic, prevent excessive queues inside the parking lot and upstream and downstream of the parking lot entrance and exit, improve the problem of uneven spatial distribution of vehicle departure and entry traffic, speed up vehicle departure and entry, and improve regional traffic efficiency.

[0067] (2) The present invention can perform adaptive adjustments based on real-time road congestion conditions. When the traffic system is impacted, it can quickly return to normal levels, thereby expanding the space available for flexible adjustment of the system.

[0068] (3) After the present invention completes the setting of the parking lot entrance and exit, the entrance and exit information can be updated in real time on the electronic display screen in the parking lot, the variable information board on the road side, and the vehicle-side navigation application to achieve accurate navigation.

[0069] (4) The present invention distinguishes between parking lots with sufficient historical data and parking lots lacking historical data to predict the number of vehicles leaving the parking lot and the number of vehicles entering the parking lot per unit time. For parking lots with historical data of more than one year, a machine learning algorithm is used to improve the prediction accuracy. For parking lots with historical data of less than one year, data from similar parking lots can be used to simply estimate the number of vehicles leaving the parking lot and the number of vehicles entering the parking lot per unit time, thereby ensuring the efficiency of system operation.

[0070] (5) The present invention is based on a deep reinforcement learning algorithm. When applying a multi-agent reinforcement learning algorithm, a performer network can be used to select appropriate actions based on the current state, and a critic network can estimate the expected benefits of the action. The combination of the performer network and the critic network reduces the training difficulty of the deep reinforcement learning model. A large amount of historical data can help the agent find a control strategy with good generalization ability. In addition, the idea of ​​"centralized training and distributed execution" allows each entrance and exit to be set according to the queue information (local information) within the parking lot and upstream and downstream, and can also achieve coordination between different parking lot entrances and exits, thereby improving the operating efficiency of the entire parking system and surrounding roads. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 is a flow chart for calculating the minimum number of lanes at the entrance and exit of a parking lot in the existing method.

[0072] Figure 2 is a diagram of the queuing theory verification

[0073] Figure 3 is an example of the principles for setting up entrances and exits when the parking lot entrances and exits are connected to the road.

[0074] FIG4 is a flow chart of the present invention

[0075] Figure 5 is the structure diagram of the artificial neural network

[0076] Figure 6 is a diagram of the long short-term memory network structure

[0077] Figure 7 is a random forest structure diagram

[0078] Figure 8 is an example of the principles for setting up entrances and exits when a parking lot entrance and exit is connected to a one-way road.

[0079] Figure 9 is an example of the principles for setting up entrances and exits when the parking lot entrance and exit are connected to a two-way road.

[0080] Figure 10 is an example of the principles for setting up entrances and exits when the entrances and exits of a parking lot are located on two roads at an intersection.

[0081] Figure 11 is an example of the entrance and exit setting principles when the parking lot entrances and exits are distributed on three roads at two intersections.

[0082] Figure 12 shows an example of the principles for setting up entrances and exits when the entrances and exits of a parking lot are distributed on three roads at two intersections.

[0083] Figure 13 is an example of the entrance and exit setting principles when the parking lot entrances and exits are distributed on four roads at four intersections DETAILED DESCRIPTION

[0084] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0085] The preparatory work of the present invention is the design and layout of the traffic information collection system and traffic information release.

[0086] The traffic information collection system relies on data from two types of sensors: video cameras and millimeter-wave radars to collect vehicle information. For the exterior and interior sections of the parking lot, the video cameras are directional cameras, and the millimeter-wave radars are long-range radars operating in the 79GHz frequency band. To align their world coordinates, the video cameras and millimeter-wave radars can be deployed at the same location on roadside poles outside the parking lot or on pillars inside the parking lot. The specific installation parameters of both sensors can be adjusted based on the required field of view. For a field of view of 100-150m, the sensors can be mounted at a height of 6 meters, with a downward tilt angle of 10°, and data collected at a frequency of 25Hz. The video cameras at the parking lot entrance and exit gates should be installed approximately 0.5m in front of the gate, at a height of 1.5m and a downward tilt angle of 20°. Data collection begins when a vehicle is detected.

[0087] The traffic information dissemination system utilizes electronic display screens, roadside variable information boards, and a vehicle-based navigation application to announce parking lot entrance and exit lane configurations. Specifically, a suspended electronic display screen is installed above the lanes within the parking lot, while wall-mounted electronic display screens are installed on the parking lot's pillars. Roadside variable information boards are installed on the roads outside the parking lot, and a navigation application that updates synchronously with the parking lot entrance and exit lane configurations is installed on vehicles that support vehicle-to-road communication. The electronic display screen within the parking lot receives route planning information provided by the vehicle's terminal and displays the vehicle's turn at the intersection and the distance to the parking lot exit on the electronic display screen. Roadside variable information boards installed on the roads outside the parking lot display the updated entrance and exit lane configuration and indicate the number of available parking spaces. The vehicle-based navigation application is used to recommend parking lot entrances and exits based on the driver's destination, using methods such as dynamic programming to plan the optimal driving route. The driver is informed of changes to parking lot entrances and exits and the planned route through voice prompts.

[0088] The present invention relates to a method for controlling variable parking lot entrances and exits. The overall technical approach is shown in FIG4 . The main steps include: receiving historical vehicle data at parking lot entrance and exit barriers; predicting the number of vehicles leaving and entering the parking lot per unit time; calculating the minimum number of required parking lot exit lanes and entrance lanes; training a parking lot entrance and exit lane configuration model; and applying the parking lot entrance and exit lane configuration model. Optionally, when applying the parking lot entrance and exit lane configuration model, internal and / or external guidance from the parking lot is used to guide drivers along a re-planned route.

[0089] The specific implementation methods are as follows:

[0090] a) Receive historical vehicle data at the parking lot entrance and exit gates

[0091] At the gate of the parking lot entrance i, the camera and vehicle target recognition and feature extraction technology are used to obtain the information of vehicle v, including the license plate number Appearance features (including model, color, etc.) Detection time And record the detection time (Include specific day of the week and time period ) and weather information Calculate parking costs Then all the vehicle data collected at the gate is recorded as D p Among them, the historical data of vehicle information collected by millimeter wave radar and camera is represented by D p,his , real-time data is represented by D p,real Parking fee Information is based on the time when the parking lot entrance and exit gate detects the vehicle entering the parking lot and time to exit the parking lot Combined with parking fee calculation function The parking fee for vehicle v is calculated as

[0092] b) Prediction of the number of vehicles leaving the parking lot per unit time and the number of vehicles entering the parking lot per unit time

[0093] For parking lots with more than one year of historical data, the historical data of vehicle departure and arrival in the parking lot are used. p,his For parking lot entrance and exit i, the detection data of vehicle v specifically includes the license plate number Appearance features (including model, color, etc.) Detection time Parking fees Information and detection time Divided into specific weeks and time period (morning peak, evening peak, flat peak), plus weather information at the time of detection Based on the data of vehicles entering and leaving the parking lot at all entrances and exits, the number of vehicles entering and leaving the parking lot at different times of the week is counted, as shown in Figure 5. The number of vehicles leaving the parking lot per unit time V is predicted using an artificial neural network. out,fu , the number of vehicles entering the parking lot per unit time V in,fu . V out,fu =f out (T,W,X,E) V in,fu =f in (T,W,X,E)

[0094] In the formula, out means leaving the parking lot, in means entering the parking lot, and f out and f in is the prediction function for the number of motor vehicles entering and leaving the parking lot, T is the code of the prediction period, that is, 1 is the morning peak, 2 is the evening peak, and 3 is the off-peak, W is the weather code, X is the day of the week, that is, Monday to Sunday are coded from 1 to 7, and E is the parking fee during the prediction period.

[0095] Variant A: As shown in Figure 6, a long short-term memory network is used to predict the number of vehicles leaving the parking lot per unit time V out,fu , the number of vehicles entering the parking lot per unit time V in,fu , using a recurrent neural network based on long-term historical parking data to predict the number of vehicles leaving the parking lot per unit time V in the future out,fu , the number of vehicles entering the parking lot per unit time V in,fuPerform inference and delete or add information to the network using a "gate" mechanism, so that the network can make predictions without the gradient vanishing and exploding problems.

[0096] Variant B: As shown in Figure 7, random forest is used to predict the number of vehicles leaving the parking lot per unit time V out,fu , the number of vehicles entering the parking lot per unit time V in,fu , using the Bootstrap resampling method to extract the historical parking data D p,his Multiple samples are extracted from the dataset and a decision tree model is built for each sample. Finally, the prediction results of multiple decision trees are combined to achieve a robust prediction of the number of vehicles leaving and entering the parking lot.

[0097] For parking lots with less than one year of historical data, based on the four-stage method, the historical data of completed parking lots that meet the land use similarity requirements are referenced and multiplied by the empirical coefficient to determine the number of vehicles leaving the parking lot per unit time and the number of vehicles entering the parking lot per unit time. Among them, the area occupied by each land use type of the parking lot space is determined. The area occupied by a certain land use type k of the current parking lot p is The land use type k of the reference built parking lot f covers an area of The ratio of the two It is defined as the similarity of land use properties. For example, if all land use types of the current parking lot and the reference parking lot meet the threshold requirement of 0.8≤δ≤1.2, then the current parking lot and the reference parking lot are considered similar. Referring to the historical data of the parking lot, a machine learning method is used to predict the number of vehicles leaving the parking lot per unit time and the number of vehicles entering the parking lot per unit time. The predicted result is multiplied by the empirical coefficient λ to determine the parking demand V of the parking lot. in,fu and V out,fu The empirical coefficient λ can be the similarity of a certain land use type or the mean of the similarities of all land use types.

[0098] c) Calculation of the minimum number of required parking lot exit lanes and entrance lanes

[0099] Based on the parking demand forecast of the number of vehicles leaving the parking lot per unit time and the number of vehicles entering the parking lot per unit time, combined with the lifting and lowering time and speed of the parking lot entrance and exit gates, the traffic capacity of an entrance and exit lane is calculated, that is, the number of vehicles that can enter or leave through a lane per hour, and then the minimum number of exit lanes and the minimum number of entrance lanes of the parking lot that meet the parking demand are calculated to ensure that when the parking lot entrance and exit lane settings change, at least the demand for vehicles leaving and entering the parking lot can be met.

[0100] Among them, according to the predicted number of vehicles leaving the parking lot per unit time V out,fu, the number of vehicles entering the parking lot per unit time V in,fu , calculate the minimum number of exit lanes and the minimum number of entrance lanes required for the parking lot:

[0101] Where c out and c in are the minimum number of exit lanes and entrance lanes required for the parking lot, respectively. [*] indicates rounding up. V out,fu and V in,fu are the predicted number of vehicles leaving the parking lot per unit time and the number of vehicles entering the parking lot per unit time, C out and C in The traffic capacity of one lane of the parking exit lane and the entrance lane can be obtained according to the calculation method of road traffic capacity:

[0102] Where C represents C out or C in , t0 is the headway (s), v is the speed (km / h), and l0 is the minimum distance between vehicles (m). t0, v, and l0 can all be obtained by taking the average of the measured vehicle speed, headway, and distance at the parking lot entrance and exit gates.

[0103] d) Parking lot entrance and exit setting model training

[0104] In traffic simulation software such as SUMO and VISSIM, a road and parking space simulation environment is established based on the actual road traffic network, parking lot entrance and exit locations, parking space internal road sections, and parking space layout. At the same time, Synchro signal timing optimization software is used to solve the optimal signal timing plan for each intersection in the entire road network and configure it in the traffic simulation software. Then, based on the predicted number of vehicles leaving the parking lot per unit time V out,fu , the number of vehicles entering the parking lot per unit time V in,fu The system randomly generates starting and ending points for each driver's trip. Using the shortest path method, the system aims to minimize the spatial path length to solve the driver's route selection plan. Specifically, departing vehicles start at the parking lot and end at a location other than the parking lot. Incoming vehicles end at the parking lot and start at a location other than the parking lot. Interaction logic and interfaces are established between the simulation software and the backend control program, allowing the backend program to obtain and control the status of road traffic facilities within the traffic simulation environment.

[0105] During traffic simulation, the backend control program captures real-time queue information for sections upstream and downstream of parking lot entrances and exits, as well as information about sections within the parking lot, the number of vehicles leaving the parking lot per unit time, and the number of vehicles entering the parking lot per unit time. Based on monitoring information about sections outside and within the parking lot, the traffic simulation platform utilizes a parking lot entrance and exit lane configuration model to optimize the lane configuration in real time. This configuration is controlled through the interface between the simulation software and the backend control program.

[0106] Variant A: Based on swarm intelligence algorithms (such as ant colony algorithms, genetic algorithms, simulated annealing algorithms, etc.), the shortest spatial path is used as the initial solution, and the shortest travel time is used as the objective function. The driving path is obtained through algorithm iteration.

[0107] In the traffic simulation platform, the queue information of the upstream road section of the parking lot entrance and exit is converted into up,sim and the queue information of the road section downstream of the parking lot entrance and exit q down,sim , parking lot internal road section queue information park,sim , Parking lot departure time V out,sim , the number of vehicles entering the parking lot per unit time V in,sim Set it as the state of the parking lot entrance and exit, and establish constraints based on the minimum number of exit lanes and entrance lanes required for the parking lot and the parking lot entrance and exit setting principles, and use the deep reinforcement learning algorithm to build a parking lot entrance and exit setting model.

[0108] To ensure that parking lot entrance and exit lanes meet practical application requirements, the adaptive entrance and exit lane configuration assumes that all lanes are flexible, allowing them to be closed or open, and when open, lanes can be freely selected as exit or entrance lanes. Each parking lot entrance and exit is an intelligent entity that can independently configure entrance and exit lanes based on its state. Taking parking lot entrance and exit i as an example, the state and entrance and exit lane configuration of the parking lot entrance and exit can be expressed as: a i (t) = {0, 1, 2, 3}

[0109] Where s i is the observed state of parking lot entrance i, is the queue information of the upstream section of the parking lot entrance and exit i, is the queue information of the downstream road section of parking lot entrance and exit i, a iis the entrance and exit lane setting of parking lot entrance and exit i, 0 is closed entrance and exit, 1 is open entrance and exit with both lanes as entrance lanes, 2 is open entrance and exit with both lanes as exit lanes, and 3 is open entrance and exit with both lanes as entrance and exit lanes. Among them, the parking lot entrance and exit lane setting scheme output by the agent needs to be implemented on the basis of meeting the entrance and exit lane setting principles. If the parking lot entrance and exit lane setting scheme meets the entrance and exit lane setting principles, the number of entrance and exit lanes of all parking lot entrances and exits is counted to obtain the statistical number. If the statistical number is lower than the minimum number of exit lanes required by the parking lot, c out and the minimum number of entrance channels c in , then, based on the principle of entrance and exit setting, randomly open exits and / or entrances, and ensure that the number of exits and entrances meets the minimum requirements. Otherwise, the entrances and exits are arranged according to the principle of entrance and exit setting. The principle of entrance and exit setting can be specifically designed based on the connection between the parking lot entrance and exit and the surrounding roads. Specific examples are shown in Figures 8 to 13.

[0110] For a parking lot, it is necessary to coordinate the setting of multiple entrances and exits to effectively solve the congestion problem caused by the uneven distribution of traffic flow in space. The overall state of the parking lot entrances and exits is x = [s1,...,s N ], the overall control method is a=[a1,...,a N The selected entrance and exit lane configuration is continuously evolved over the subsequent evaluation period (Δt), using the system's built-in vehicle following model, lane changing model, and real-time path planning model. After Δt, multi-agent deep reinforcement learning evaluates the effectiveness of the entrance and exit lane configuration using a reward function.

[0111] The reward function is used to evaluate the impact of the entrance and exit configuration on dynamic and static traffic. Considering the collaborative relationship between the various entrances and exits of the parking lot, the goal of the parking lot entrance and exit configuration is to achieve coordination among all entrances and exits of the entire parking lot while meeting the needs of vehicle departure and arrival, thereby reducing the length of queues within the parking lot and surrounding roads and improving the efficiency of vehicle departure and arrival. The overall reward function can be expressed using the average queue length of the road section within time Δt:

[0112] Where w1, w2, w3, w4, and w5 are weights. and are the average queue lengths of the upstream and downstream sections of entrance and exit i within time Δt, is the average queue length in the parking lot during time Δt.

[0113] Considering that the entrance and exit setting scheme needs to meet the entrance and exit setting principles and try to ensure that no traffic conflicts occur, if the output scheme a does not meet the entrance and exit setting principles, the agent will be punished, that is, r e =-100; if the entrance and exit setting principle is met, then r e =0.

[0114] In summary, the reward function is: r = r q +r e

[0115] Based on the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) network, we completed the training of the parking lot entrance and exit lane setting model. As the simulation progressed, we collected a large amount of interactive data sent back in real time by the traffic simulation platform.<x,a,r,x'> , and store it in the experience pool. Based on the idea of ​​"centralized training", the global entrance and exit setting strategy is optimized by utilizing the actions and states of the current agent and other agents. According to the number of entrances and exits of the parking lot, a corresponding critic network (Critic) and performer network (Actor) are established. Each critic network and performer network consists of an Evaluate network and a Target network. The critic network is used to learn how to evaluate the collaborative control strategy, and the performer network is used to determine the settings of each entrance and exit of the parking lot. In particular, the setting method of each parking lot entrance and exit is controlled by a group of performer networks and evaluated by a group of critic networks. Specifically, each performer network inputs the state of the current entrance and exit and outputs an action, while the critic network outputs the evaluation Q value of the collaborative control effect of the current entrance and exit based on the state of the entire parking lot and all actions. When the reward function shows a convergence trend as the simulation progresses, the model completes training and obtains model M1. During the training process, the loss function of the critic network is:

[0116] Where μ i is the strategy function represented by the Evaluate network in the performer network of parking lot entrance i, S is the total number of samples extracted from the experience pool for model training, <x j ,a j ,r j ,x' j > is a sample drawn from the experience pool, is the critic network corresponding to the parking lot entrance i and the policy function μ, is the entrance and exit setting output by the Evaluate network in the performer network of the parking lot entrance and exit k in the jth sample, N is the total number of entrances and exits of the parking lot, γ is the discount rate, oj is the state of the agent observed by the simulation platform, μ' k is the policy function represented by the Target network in the performer network at the parking lot entrance k, It is the control strategy output by the Target network in the performer network of the parking lot entrance and exit k in the jth sampling data.

[0117] The policy gradient function used when updating the performer network is:

[0118] Where θ i are the parameters of the Evaluate network in the critic network for parking lot entrance and exit i. In particular, parking lot entrances and exits are uniquely numbered, and i and k are used here to distinguish the Evaluate network from the Target network.

[0119] For parking lot entrance and exit i, the update method of its Target network is: θ' i ←τθ i +(1-τ)θ' i μ' i ←τμ i +(1-τ)μ' i

[0120] For a parking lot, the total number of entrances and exits is limited. Therefore, in order to reduce the difficulty of network training, the performer and critic networks of different agents can adopt the same structure. For example, for a parking lot with a total of 4 entrances and exits and each entrance and exit is set as a double lane, the number of neurons in each layer of the performer network is 5-50-30-20-1, and the number of neurons in each layer of the critic network is 15-50-30-20-1. The activation of the middle layer of the neural network uses relu. The last layer of the performer network uses a sigmoid function and multiplies the output result by 3 to map it to the preset action space, and at the same time takes the integer part of the output result. In particular, for the variable representing the state of the entire parking lot in each parking lot entrance and exit state, that is, the queuing information q of the internal road section of the parking lot park,sim , Parking lot departure time V out,sim , the number of vehicles entering the parking lot per unit time V in,sim , in order to avoid duplication, it is only entered once in the neural network. The network parameters are set as shown in the following table:

[0121] Table 3 Neural network parameter settings

[0122] The parking lot entrance and exit lane configuration model is fed with queue lengths upstream and downstream of parking lot entrances and exits, queue lengths within the parking lot, the number of vehicles leaving the parking lot per unit time, and the number of vehicles entering the parking lot per unit time. The vehicle-road cyber-physical system then outputs the configuration of each parking lot entrance and exit, completing the parking lot entrance and exit lane configuration. Specifically, in the model's practical application, a "distributed execution" approach is employed, with each entrance and exit utilizing its own performer network and local state information to make decisions about entrance and exit lane configuration, enabling rapid responsiveness of the parking system. Once the vehicle-road cyber-physical system receives information about the changed parking lot entrance and exit lane configuration, it immediately transmits it to vehicles and roadside variable information devices, alerting drivers to the changes. These variable information devices include electronic display screens within the parking lot and variable roadside signage. In addition to providing an overview of the parking lot entrance and exit lane configuration, the electronic display screens within the parking lot also provide vehicle direction guidance.

[0123] Variant A: Based on the Deep Q network, the parking lot entrance and exit setting model training is completed. As the simulation continues, a large amount of interactive data is collected in real time from the traffic simulation platform.<x,a,r,x'> , and store it in the experience pool, and use the actions and states of the current agent and other agents to optimize the entrance and exit layout strategy of each parking lot. The deep Q network outputs the entrance and exit layout plan for all parking lot entrances and exits based on the overall state of the parking lot entrances and exits. Specifically, the entrance and exit layout plan selected is the one that can maximize the expected reward, that is, a=max a Q(s,a;θ)

[0124] Where Q*(x,a) is the optimal Q function, ε is the environment, and γ is the reduction coefficient.

[0125] In actual operation, the neural network Q(x,a;θ) is used to approximate Q*(x,a), where θ is the neural network parameter. During network training, the total samples used for model training are randomly extracted from the experience pool. <x j ,a j ,r j ,x' j > is the sample drawn from the experience pool, S is the total number of samples drawn from the experience pool for model training, then the loss function of the Q network can be written as y=[r j +γmax a' Q(x' j ,a' j ;θ)]

[0126] For a parking lot, the total number of entrances and exits is limited, so the action space of the network is not too large. Single-agent reinforcement learning can also solve the parking lot entrance and exit road setting scheme. For example, for a parking lot with 4 entrances and exits and each entrance and exit is set as a double lane, the number of neurons in each layer of the Q network is 11-100-50-16, and the activation of the middle layer of the neural network uses ReLU. In particular, for the variable representing the state of the entire parking lot in each parking lot entrance and exit state, that is, the queue information q of the internal road section of the parking lot park,sim , Parking lot departure time V out,sim , the number of vehicles entering the parking lot per unit time V in,sim , in order to avoid duplication, it is only entered once in the neural network. The network parameters are set as shown in the following table:

[0127] Table 4 Neural network parameter settings

[0128] Variant B: Based on the Deep Deterministic Policy Gradient (DDPG) network, the parking lot entrance and exit lane setting model training is completed. As the simulation continues, a large amount of interactive data is collected in real time from the traffic simulation platform.<x,a,r,x'> , and store it in the experience pool, regard the vehicle-road information-physical system as an intelligent agent, and use the current intelligent agent's actions and states to optimize the global entrance and exit setting strategy. Establish two critic networks (Critic) and performer networks (Actor), both of which are composed of Evaluate network and Target network. The critic network is used to learn how to evaluate the parking lot entrance and exit setting strategy, and the performer network is used to determine the settings of each parking lot entrance and exit. In particular, the setting method of all parking lot entrances and exits is controlled by a group of performer networks and evaluated by a group of critic networks. Specifically, the performer network inputs the status of all parking lot entrances and exits and outputs actions, while the critic network outputs the evaluation Q value of the collaborative control effect of the current entrance and exit based on the status of the entire parking lot and all actions. When the reward shows a convergence trend as the simulation progresses, the model has completed training and the model M1 is obtained. During the training process, the loss function of the critic network is:

[0129] In the formula, S is the total number of samples extracted from the experience pool for model training, r t As reward, Q'(x t ,μ'(x t |θ μ' )|θ Q' ) is the Target network in the critic network set up at the entrance and exit of the parking lot, Q(x t ,a t |θQ ) is the Evaluate network in the critic network set up at the entrance and exit of the parking lot, μ'(x t |θ μ' ) is the Target network in the performer network set up at the entrance and exit of the parking lot, θ μ' The parameters of the Target network in the performer network set for the parking lot entrance and exit, θ Q' ,θ Q Parameters of the Target network and the Evaluate network in the critic network set for the parking lot entrance and exit.

[0130] The policy gradient function used when updating the performer network is:

[0131] Where θ μ Parameters of the Evaluate network in the Performer network set for the parking lot entrance and exit.

[0132] For the parking lot entrance and exit, the update method of the Target network is: θ'←τθ+(1-τ)θ' μ'←τμ+(1-τ)μ'

[0133] For a parking lot, the total number of entrances and exits is limited. Therefore, in order to reduce the difficulty of network training, the performer and critic networks of different intelligent agents can adopt the same structure. For example, for a parking lot with a total of 4 entrances and exits and each entrance and exit is set as a double lane, the number of neurons in each layer of the performer network is 11-50-30-20-4, and the number of neurons in each layer of the critic network is 15-50-30-20-1. The activation of the middle layer of the neural network uses relu. The output layer of the performer network uses a sigmoid function and multiplies the output result by 3 to map it to the pre-set action space, and at the same time takes the integer part of the output result. In particular, for the variable representing the state of the entire parking lot in each parking lot entrance and exit state, that is, the queuing information q of the internal road section of the parking lot park,sim , Parking lot departure time V out,sim , the number of vehicles entering the parking lot per unit time V in,sim , in order to avoid duplication, it is only entered once in the neural network. The network parameters are set as shown in the following table:

[0134] Table 3 Neural network parameter settings

[0135] e) Application of parking lot entrance and exit model

[0136] On road section R, millimeter-wave radar is used to obtain vehicle data within the radar field of view. Can include world geographic coordinates Speed ​​information Detection time At the same time, the camera is used to obtain vehicle video image data within the camera field of view Including the vehicle license plate number obtained based on vehicle target recognition and feature extraction technology Appearance features (including model, color, etc.) Positioning coordinates Detection time Lane Based on the positioning coordinates of the millimeter wave radar and the camera, the positioning coordinates of the vehicle are converted into a specific lane and a specific position on the lane, and then the information of the vehicle v (including the license plate number) is converted into a specific lane and a specific position on the lane. Appearance characteristics Lane coordinate speed ) and detection time Integrate and record together, and further extract the queue information of the upstream section of the parking lot entrance and exit according to the vehicle speed, lane and detection time up,real , Queue information of the road section downstream of the parking lot entrance and exit q down,real , parking lot internal road section queue information park,real At the same time, the real-time detection data D collected by the camera at the entrance and exit of the parking lot p,real In the example, based on the number of vehicles entering and exiting the parking lot in the past hour, the real-time number of vehicles leaving the parking lot per unit time V is obtained. out,real , the number of vehicles entering the parking lot per unit time V in,real .

[0137] Real-time road traffic information is collected by two types of sensors: video cameras and millimeter-wave radars. In the road section R inside or outside the parking lot, the millimeter-wave radar is used to obtain vehicle data within the radar field of view. Can include world geographic coordinates Speed ​​information Detection time At the same time, the camera is used to obtain vehicle video image data within the camera field of view Including the vehicle license plate number obtained based on vehicle target recognition and feature extraction technology Appearance features (including model, color, etc.) Positioning coordinates Detection time Lane Based on the positioning coordinates of the millimeter wave radar and camera, the vehicle's positioning coordinates are converted into a specific lane and a specific position on the lane, and then the vehicle v information is integrated and recorded as speed. Lane coordinate license plate number Appearance characteristics (including model, color, etc.) data Detection time The data format is shown in Table 1. By using data fusion technology, the traffic data of each road section are integrated to obtain the real-time parking lot entrance and exit upstream road section queue information q up,real , Queue information of the road section downstream of the parking lot entrance and exit q down,real , parking lot internal road section queue information park,real .

[0138] Table 1 Real-time road traffic information of the external and internal sections of the parking lot

[0139] Use the camera at the parking lot entrance and exit i gate to capture the vehicle appearance and license plate, use support vector machine, convolutional neural network, generative adversarial network and other machine learning algorithms to identify vehicle features and license plate information, and obtain the vehicle license plate number Appearance features (including model, color, etc.) Detection time And record the detection time (Include specific day of the week and time period ) and weather information Calculate parking costs Composition of vehicle data D p,real Real-time detection data collected from the camera at the entrance and exit of the parking lot D p,real In the example, based on the number of vehicles entering and exiting the parking lot in the past hour, the real-time number of vehicles leaving the parking lot per unit time V is obtained. out,real , the number of vehicles entering the parking lot per unit time V in,real .

[0140] Table 2 Real-time road traffic information for entering and leaving parking lot entrance and exit i

[0141] When setting up the application at the entrance and exit of the parking lot, the millimeter-wave radar and camera are input into the trained model M1 to detect the real-time queue information q of the upstream section of the parking lot entrance and exit. up,real , Queue information of the road section downstream of the parking lot entrance and exit q down,real , parking lot internal road section queue information park,real , the number of vehicles leaving the parking lot per unit time V out,real , the number of vehicles entering the parking lot per unit time V in,real , each entrance and exit of the output parking lot is equipped with a real .

[0142] f) Parking lot internal and / or external guidance (optional)

[0143] If the parking lot entrance and exit lane layout changes, vehicles inside the parking lot can, on the one hand, communicate in real time with parking lot facilities via vehicle-to-road communication equipment within the parking lot, using indoor positioning and communication technologies such as WLAN, RFID, UWB, and Bluetooth, to obtain the updated parking lot exit lane layout. This allows the vehicle-based navigation application to replan its driving route, allowing the driver to use the vehicle-based navigation application to plan the shortest driving route in real time. Furthermore, wall-mounted and suspended electronic displays within the parking lot provide real-time vehicle direction information, enabling real-time navigation within the parking lot. Simultaneously, a map application loads indoor panoramic and local maps of the parking lot, displaying the planned route and heading, to help the driver find the parking lot exit. If a vehicle outside the parking lot needs to enter the parking lot after the entrance and exit lane layout changes, the driver can obtain the updated entrance and exit lane layout and the number of available parking spaces from a roadside variable information board. Alternatively, the driver can communicate with roadside smart infrastructure using communication technologies to obtain this information and load it into a high-precision map application, enabling the vehicle-based navigation application to replan its driving route. The vehicle then proceeds to the corresponding parking lot entrance based on the navigation information displayed on the map application.

[0144] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

Claims

1. A method for adaptively controlling variable parking lot entrances and exits, the method involving parking lot exterior sections and parking lot interior sections; the exterior sections include sections upstream and downstream of the parking lot entrance and exit; the interior sections include sections where vehicles travel within the parking lot and sections where vehicles queue and / or travel at the parking lot entrance and exit; the method comprising the following steps: a) Receive historical vehicle data D at the parking lot entrance and exit gates p ; b) Based on historical vehicle data D at the parking lot entrance and exit gates p , predict the number of vehicles leaving the parking lot per unit time V out,fu and the number of vehicles entering the parking lot per unit time V in,fu ; c) Based on the predicted number of vehicles leaving the parking lot per unit time V out,fu , the number of vehicles entering the parking lot per unit time V in,fu Calculate the minimum number of parking lot exit lanes c required out and the minimum number of entrance channels c in ; d) In traffic simulation, the queue information of the upstream road section of the parking lot entrance and exit transmitted back by the simulation platform in real time up,sim , Queue information of the road section downstream of the parking lot entrance and exit q down,sim , parking lot internal road section queue information park,sim , the number of vehicles leaving the parking lot per unit time V out,sim , the number of vehicles entering the parking lot per unit time V in,sim As state x, the minimum number of parking lot exit lanes c required out and the minimum number of entrance channels c in , parking lot entrance and exit setting principles as constraints, build parking lot entrance and exit setting model M0, output parking lot entrance and exit setting plan a, and then implement the plan at the parking lot entrance and exit in the simulation platform, obtain the parking lot entrance and exit state x' at the next moment, use reward r to evaluate plan a, and use interactive data<x,a,r,x'> The parking lot entrance and exit road setting model is trained until convergence, and a trained parking lot entrance and exit road setting model M1 is obtained; e) After the model training is completed at the parking lot entrance and exit, the real-time parking lot entrance and exit upstream section queue information q is input into the trained model M1 up,real , Queue information of the road section downstream of the parking lot entrance and exit q down,real , parking lot internal road section queue information park,real , the number of vehicles leaving the parking lot per unit time V out,real , the number of vehicles entering the parking lot per unit time V in,real , output parking lot entrance and exit setting plan a real .

2. The adaptive parking lot variable entrance and exit control method according to claim 1, characterized in that: If the parking lot has historical parking data for more than one year, the number of vehicles leaving the parking lot per unit time V out,fu , the number of vehicles entering the parking lot per unit time V in,fu Determined based on historical data using a machine learning algorithm; if the parking lot has less than one year of historical parking data, the four-stage method is used to predict the similarity of the land use properties of the ground above different underground parking lots using the land use similarity δ. The historical data of completed parking lots with similar land use properties are then referenced and multiplied by the empirical coefficient λ to predict the number of vehicles leaving the parking lot per unit time V. out,fu , the number of vehicles entering the parking lot per unit time V in,fu .

3. The adaptive parking lot variable entrance and exit control method according to claim 1, characterized in that: The adjustment process of the parking lot entrance and exit road setting under the constraint condition is as follows: after the parking lot entrance and exit road setting plan a is output by the parking lot entrance and exit road setting model, if the entrance and exit road setting plan a meets the entrance and exit road setting principle, the number of exit roads and entrance roads of all parking lot entrances and exits is counted to obtain the statistical number, and if the statistical number is lower than the minimum number of exit roads c required by the parking lot, out and the minimum number of entrance channels c in , then the exits and / or entrances are randomly opened on the basis of meeting the entrance and exit setting principles, and the number of exits and entrances meets the minimum requirements of each; otherwise, the entrances and exits are arranged according to the entrance and exit setting principles.

4. The adaptive parking lot variable entrance and exit control method according to claim 1, characterized in that: The parking lot entrance and exit road is set up in accordance with the principle that the traffic entering and exiting the parking lot does not intersect with the traffic on the road, and vehicles are allowed to turn right to enter and exit the parking lot.

5. The adaptive parking lot variable entrance and exit control method according to claim 1, characterized in that: The reward r of the parking lot entrance and exit road setting model is only used in the model training process, and is represented by the average queue length of the road section, vehicle driving efficiency, and whether the parking lot entrance and exit road setting principles are met; among them, the reward related to the average queue length of the road section is the average queue length of the upstream section of the parking lot entrance and exit in the traffic simulation platform. Average queue length on the road downstream of the parking lot entrance and exit Average queue length in the parking lot interior section Rewards related to vehicle driving efficiency utilize the number of vehicles leaving the parking lot per unit time V in the traffic simulation platform. out,sim and the number of vehicles entering the parking lot per unit time V in,sim Indicates that for the solution output by the parking lot entrance and exit setting model, if it meets the parking lot entrance and exit setting principles, the reward is set to 0; if it does not meet the parking lot entrance and exit setting principles, the reward is set to a negative number.

6. The adaptive parking lot variable entrance and exit control method according to claim 1, characterized in that: The method for constructing the parking lot entrance and exit road setting model is as follows: Based on the deep reinforcement learning algorithm, a neural network is established, and the interaction data between the parking lot entrance and exit road setting model M0 and the traffic simulation environment is used to return in real time the traffic simulation platform.<x,a,r,x'> , iteratively train the network to obtain the model [M0,M0',...], and after the model training converges, the parking lot entrance and exit setting model M1 is obtained.

7. The adaptive parking lot variable entrance and exit control method according to claim 1, characterized in that: The application method of the parking lot entrance and exit road setting model is as follows: When the parking lot entrance and exit road setting model is applied to the parking lot entrance and exit, the real-time parking lot entrance and exit upstream road section queue information q up,real , Queue information of the road section downstream of the parking lot entrance and exit q down,real , parking lot internal road section queue information park,real , the number of vehicles leaving the parking lot per unit time V out,real , the number of vehicles entering the parking lot per unit time V in,real As the state, input the parking lot entrance and exit setting model M1, and output the parking lot entrance and exit setting plan a real , change the exit and entrance settings of the parking lot entrance and exit lanes.

8. The adaptive parking lot variable entrance and exit control method according to claim 1, characterized in that: The parking lot entrance and exit has at least one lane, and the parking lot entrance and exit settings are to set the lane as an exit lane, an entrance lane or a closed lane.

9. The adaptive parking lot variable entrance and exit control method according to claim 1 further comprises the following steps: f) After the parking lot entrance and exit road arrangement plan is changed, vehicles outside the parking lot that need to enter the parking lot and vehicles inside the parking lot that need to leave the parking lot are guided.

10. The adaptive parking lot variable entrance and exit control method according to claim 9, characterized in that: There is at least one variable information board on the road side of the parking lot outside the road section, and the driver obtains the updated parking lot entrance and exit road setting plan from the variable information board on the road side. real and the number of available parking spaces.

11. The adaptive parking lot variable entrance and exit control method according to claim 9, characterized in that: The internal road section of the parking lot also includes at least one electronic display screen. According to the updated parking lot entrance and exit road setting plan a real , using the electronic display screen in the parking lot to prompt the vehicle's driving direction in real time, realize real-time navigation in the parking lot, and help drivers find the parking lot exit.

12. The adaptive parking lot variable entrance and exit control method according to claim 9, characterized in that: The vehicle has a vehicle-road communication function and a navigation application, including at least one of a mobile navigation application and a vehicle-side navigation application, and can obtain an updated parking lot entrance and exit setting plan through vehicle-road communication. real And load the high-precision map application to re-plan the driving route, so that the vehicle can go to the corresponding parking lot entrance and exit according to the navigation information on the map application.

13. An adaptive parking lot variable entrance and exit control device, comprising a memory, a processor, and a program stored in the memory, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 12 is implemented.

14. A storage medium having a program stored thereon, characterized in that: When the program is executed, the method according to any one of claims 1 to 12 is implemented.