A method for adaptive control of variable entrances and exits in a parking lot

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

Description

Technical field The invention relates to the field of parking facility control technology, specifically to a method for adaptive control of variable entrances and exits in a parking lot based on traffic state perception and intelligent management. Background With the development of the automotive industry, the increasing demand for motor vehicle travel has not only put pressure on road traffic but also made parking system management more challenging. A large urban parking lot often have multiple entrances and exits, all following die same parking lot entrance and exit lane setup plan. This can lead to situations where some entrances and exits are congested while others are underutilized. Effectively managing parking lot entrance and exit lane setup plans becomes a significant challenge for traffic management and control during peak times. The number of entrance lanes and the number of exit lanes in a parking lot determines whether traffic demand can be met, and the distribution of these lanes affects how efficiently vehicles can enter or exit the parking lot. Improper lane setup can not only degrade user experience but also place significant pressure on external traffics the parking lot. Although existing methods for setting parking lot entrance and exit lanes can generate lane setups are based on factors such as traffic volume, parking lot size, and road grade at entrances and exits, the traffic system is dynamic and often exhibits noticeable tidal patterns. If parking lot entrance and exit lane setup plans are based solely on limited and fixed traffic information, they may fail to meet the demand for vehicle departures and arrivals, leading to inefficiencies and even traffic gridlock. Moreover, as connection points between dynamic and static traffic, traditional fixed parking lot entrance and exit lane setup plans lack adaptability. After a disruption in the traffic system, these setups struggle to quickly return to normal levels, leaving very limited room for flexible adjustment. Currently, existing methods for setting parking lot entrance and exit lanes mainly include: 1. Parking Lot Entrance and Exit Lane Setup Method Based on Parking Space Quantity According to the "Standards for Traffic Design and Parking Garage (Lot) Setup in Construction Projects" (DGJ08-7-2006), the following guidelines for generating parking lot entrance and exit lane setup plans based on the number of parking spaces are proposed: (1) When the number of parking spaces is less than 100, the parking garage should have at least one entrance or exit with two lanes or two single lanes for an entrance and exit. (2) When the number of parking spaces is greater than or equal to 100 and less than 200, there should be at least two single-lane for an entrance and exit. (3) When the number of parking spaces is greater than or equal to 200 and less than 700, the parking facility should have at least two lanes for an entrance and two lanes for an exit. (4) When the number of parking spaces is 700 or more, the facility should have at least three two-lane entrances and exits, and a service level evaluation should be conducted to determine the appropriate number of entrances and exits. 2. Parking Lot Entrance and Exit Lane Setup Method Based on Queuing Theory Assuming that the traffic flow entering and exiting the parking lot follows a Poisson distribution, and the 1 time for vehicles to pass through barrier gates follows an exponential distribution, the total number of barrier gates can be considered equivalent to the number of service stations. The time vehicles spend parking can be considered infinite compared to the gate passing time. Therefore, the processes of entering and exiting the parking lot can be modeled using queuing theory. For a parking plot with multiple entrances and exits, the queuing system can be viewed as consisting of multiple M / M / l queues or an M / M / S system. In practical applications, you need to calculate the peak hour arrival rate of vehicles and the average service time of the parking lot entrance and exit barrier gates. Based on these calculations, the minimum number of entrance lanes and the minimum number of exit lane can be determined. 3. Parking Lot Entrance and Exit Lane Setup Method Considering Road Connectivity When a parking lot entrance and exit connect with a one-way road, parking lot entrance and exit lane setup plans should prioritize avoiding cross-traffic. Typically, this involves placing entrance lanes first along the direction of the road and then exit lanes, thereby minimizing the impact of vehicles entering and exiting the parking lot on the road traffic. When parking lot entrances and exits connect with two-way roads, the rules are to ensure that the entering and exiting traffic flows do not cross with the external traffics and that vehicles turn right to enter the parking lot. When the parking lot entrances and exits are located on different sides of an intersection, entrance lanes and exit lanes are generally arranged according to the clockwise direction of vehicle flow around the intersection. The methods for setting parking lot entrance and exit lanes are based on rules that determine the number and location of these lanes. The main process involves calculating the minimum number of entrance lanes and the minimum number of exit lanes based on the total number of parking spaces, peak hour arrival rates of vehicles, average service times at the parking lot entrance and exit barrier gates, and the connecting roads (see Figure 1). This is done using queueing theory calculations (see Figure 2), and then additional lanes are added as necessary to meet service level requirements. When setting the locations of entrance and exit lanes, it is important to consider the relationship between parking lot entrances and exits and their connecting roads (see Figure 3), while adhering to the rules above. Prior Art: CN114758497A Terms 1. Adaptive control of variable entrances and exits: the entrance and exit of a parking lot adopt a variable lane setup mode, where a lane can be designated as either exit lane, entrance lane, or closed lane. The adaptive control is carried out based on parking lot entrance and exit lane setup plans generated by a parking lot entrance and exit lane setup model. 2. Dynamic traffic information: dynamic traffic information of the parking lot refers to real-time data such as queue information of upstream road segments, queue information of downstream road segments, the number of vehicles exiting parking lot per unit time, and the number of vehicles entering parking lot per unit time. 3. Static traffic information: static traffic information of the parking lot refers to the queue information of internal road segments in the parking lot, which can be represented by average queue length of internal road segments and average driving speed on internal road segments. 4. External road segment: upstream road segments and downstream road segments connect to entrances and exits of a parking lot. 5. Upstream road segment: the road segment between a parking lot entrance and exit and its upstream intersection, bounded by the two intersections. 6. Downstream road segment: the road segment between a parking lot entrance and exit and the downstream intersection, bounded by the two intersections. 7. Internal road segment: the road segments vehicles travel within the parking lot and the road segments where vehicles queue and / or move at the parking lot entrances and exits. 8. Vehicle data on road segments: positioning coordinates, speed information, detection time, license plate number, visual characteristics (including vehicle type, color, etc.), and lane location on external and internal road segments of the parking lot. 9. Vehicle data at barrier gates: license plate number, visual characteristics (including vehicle type, color, etc.), detection time (including specific day of the week and time period), weather conditions, and parking fees at barrier gates of parking lot entrances and exits. 10. Queue information: average queue length of road segments and average vehicle speed on road segments. 11. Queue length: the number of vehicles with a speed below 0.1 m / s. 12. Number of vehicles exiting parking lot per unit time: the number of vehicles that exit a parking lot within a specified time. 13. Number of vehicles entering parking lot per unit time: the number of vehicles that enter a parking lot within a specified time. 14. Parking lot entrance and exit lane setup plan: the number and location of the parking lot entrance and exit lanes, indicating that the entrance and exit lanes are specifically set on a certain lane of a entrance or exit. 15. Minimum number of entrance lanes: the least number of entrance lanes to meet the basic demand for vehicles entering the entire parking lot. 16. Minimum number of exit lanes: the least number of exit lanes to meet the basic demand for vehicles exiting the entire parking lot. 17. Parking lot entrance and exit lane setup model: use queue information of upstream road segments, queue information of downstream road segments, queue information of internal road segments, the number of vehicles exiting parking lot per unit time, and the number of vehicles entering parking lot per unit time as inputs; The minimum number of entrance lanes and the minimum number of exit lanes, along with the rules for setting the entrance and exit lanes, serve as constraints to output parking lot entrance and exit lane setup plans. 18. Interaction data: in an established simulation platform, a simulated parking lot entrance and exit lane setup model outputs a setup plan a, based on the observed state st at the parking lot entrance and exit i . After executing the setup plan in simulation platform, the state at the parking lot entrance and exit change to st'. The data on states before and after the change and st', the setup plan at for the parking lot entrance and exit, and evaluation results r of setup plan effectiveness are recorded. Subsequently, the overall state of parking lot entrances and exits is defined as x = and the overall setup plan for parking lot entrances and exits is defined as a = [q,..., qj, which is referred to as interaction data (x,a,r,x'), where = ., 19. Experience pool: a database that stores interaction data, capable of providing interaction data collected at different times for model training. 20. Reward function: based on the overall state x and x' of the parking lot entrances and exits observed by the platform and parking lot entrance and exit lane setup plan a . the reward function evaluates the effectiveness of the setup plan a outputted by a parking lot entrance and exit lane setup model. The evaluation result serves as a reward r, which is the computed outcome of the reward function. 21. Historical vehicle data: historical records of vehicles exiting and entering a parking lot at all barrier gates, including time, entrance and exit locations, vehicle types, weather conditions, and parking fee information, etc., which is a historical vehicle data at barrier gates. 22. Machine learning algorithm: including algorithms capable of performing predictions by leveraging big data, such as random forests, support vector machines, and neural networks. 23. Four-step predicting: a traffic demand predicting method used in the transportation field, which includes four stages: trip generation, trip distribution, mode choice, and traffic assignment. 24. Land similarity: the land use types above parking lots are determined according to the ‘Urban Land Classification and Planning Construction Land Standards'; This determines the area occupied by each land use type, where the area occupied by a specific type k of land use in the current parking lot is denoted as ukp . and the area occupied by the same type k of land use in the reference completed parking lot is denoted as uk; The ratio of these two areas is defined as land use type similarity Sk ; If all land use types of the current parking lot and the reference parking lot meet the threshold criteria <S< , it is considered that the current parking lot and the reference parking lot are similar; The historical vehicle data from the reference parking lot can be used to predict the number of vehicles exiting parking lot per unit and the number of vehicles entering parking lot per unit time. 25. Empirical coefficient: based on completed parking lot that meet the threshold criteria <5ran <S <<5max for land similarity characteristics with the specific parking lot, the land similarity of a specific land use type or the average land similarity of all land use types can be used as the empirical coefficient. 26. Multi-agent deep deterministic policy gradient network: an algorithm used in deep reinforcement learning for solving collaborative decision-making among multiple agents. 27. Critic network: in the multi-agent deep deterministic policy gradient network, critic network takes the observed state and parking lot entrance and exit lane setup plans as inputs and outputs the expected benefits, used to evaluate the effectiveness of the selected parking lot entrance and exit lane setup plans in the current state. 28. Actor network: in the multi-agent deep deterministic policy gradient network, actor network takes the observed state as input and outputs the parking lot entrance and exit lane setup plans, used for setting appropriate parking lot entrance and exit lane. 29. Deep deterministic policy gradient network: an algorithm in deep reinforcement learning used for solving decision-making for a single agent; By establishing two groups of separate networks, critic networks and actor networks, it directly outputs actions along with their effectiveness evaluations, effectively addressing optimization problems with continuous action spaces. 30. Deep Q-network: an algorithm in deep reinforcement learning used for solving decision-making for a single agent. It relies solely on neural networks to evaluate the effectiveness of actions and find the optimal action. 31. Convergence: when the training of a parking lot entrance and exit lane setup model reaches a later stage, the effectiveness of parking lot entrance and exit lane setup plans is evaluated using a reward function; If obtained rewards are high and the difference between rewards in adjacent training episodes falls within a specific range, it indicates that the model has converged. 32. Electronic display screen: wall-mounted or suspended electronic screens are installed on structural columns or at the top of lanes in a parking lot; These screens dynamically display changes in the parking lot entrance and exit lane setup plans, allowing for timely adjustments to internal vehicle guidance strategies and modifications to the directions for exiting the parking lot. 33. Roadside variable message sign: electronic display screens are installed along the roads outside a parking lot; These screens, in conjunction with hierarchical guidance, inform passing vehicles within a certain radius of the parking lot about the entrance and exit lane setup plans and the number of available parking spaces. 34. Navigation application: a mobile navigation application or an on-board navigation application assists drivers that need to enter or exit a parking lot by leveraging vehicle positioning and destination information; These applications autonomously select the optimal parking lot entrance and exit and perform corresponding optimal route replanning to guide drivers in controlling their vehicles. Invention content The objective of this invention is to provide a method for adaptive control of variable entrance and exit in a parking lot. By comprehensively considering dynamic traffic information and static traffic information, the method dynamically adjusts the parking lot entrance and exit lane setup plan and guides vehicles to alter their travel routes to reach their destination via the recommended parking lot entrances and exits. Tins improves the efficiency of vehicles entering and exiting the parking lot, reduces queuing on both internal road segments and external road segments, and prevents traffic congestion caused by uneven spatial distribution of traffic flow. Overall, it enhances the operational efficiency and service level of both dynamic and static traffic. The method for adaptive control of variable entrance and exit in a parking lot dynamically adjusts parking lot entrance and exit lane setup plans, altering the number and location of entrance and exit lanes. Combined with traffic guidance measures, it guides drivers to change their driving routes. The main process is shown in Figure 4, and the primary steps include: collect historical vehicle data of entrances and exits of the parking lot, predict a number of vehicles exiting parking lot per unit time and a number of vehicles entering parking lot per unit time, calculate the minimum number of entrance lanes and the minimum number of exit lanes, train the parking lot entrance and exit lane setup model, and apply the parking lot entrance and exit lane setup model. Optionally, when applying the setup model, with guidance internal and / or external, guide drivers within and / or outside the parking lot to follow the newly planned routes. The method of this invention includes the following steps: a) Collect historical vehicle data of entrances and exits of a parking lot At the i -th entrance and exit barrier gate of the parking lot, cameras utilize vehicle recognition and feature extraction technologies to obtain information about vehicles v, including license plate numbers A', visual characteristics (such as vehicle type and color) , and detection time T’ (including specific days of the week .V' and time periods H') and weather information WJ, and calculates parking fees E'. Consequently, vehicle data at barrier gates is recorded as Dp, where the historical vehicle data captured by the cameras is represented as Dp , and the real-time vehicle data captured by the cameras is represented as Dp real. b) Predict a number of vehicles exiting parking lot per unit time and a number of vehicles entering parking lot per unit time Utilizing the historical vehicle data collected from barrier gates of the parking lot Dp his. we can predict the number of vehicles existing parking lot per unit time and the number of vehicles entering parking lot per unit time . If the parking lot has at least one year of historical vehicle data D , the number of vehicles exiting parking lot per unit time and the number of vehicles entering the parking lot per unit time VjnJh will be predicted using machine learning algorithms. If tire historical vehicle data Dp his is less than one year, the prediction of the number of vehicles exiting parking lot per unit time Voutfu and the number of vehicles entering parking lot per unit time Vjn will be based on the four-step predicting. Additionally, historical vehicle data Df from a completed parking lot f with land similarity 8 between 8, and 8a will be referenced. The number of vehicles exiting parking lot per unit time VM and the number of vehicles entering parking lot per unit time V will then be calculated by multiplying the historical vehicle number by an empirical coefficient. The empirical coefficient A of the parking lot p can be calculated based on the similarity 6 of a specific land use type or the average similarity of all land use types. c) Calculate the minimum number of entrance lanes and the minimum number of exit lanes The minimum numbers of entrance lanes a and the minimum numbers of exit lanes c , are calculated based on the prediction of the number of vehicles exiting parking lot per unit time V ,. and number of vehicles entering parking lot per unit time E; fu, ensuring that the parking lot can at least meet the demand for vehicle entrance and exit when the parking lot entrance and exit lane setup plan changes. d) Train the parking lot entrance and exit lane setup model In a simulation platform, establish a road network, parking lot entrance and exit, internal road segments, and the layout of parking spaces based on the actual road traffic network. Drivers' origin-destination points and route choices are simulated based on the predicted traffic demand. An interaction logic between the simulation software and the backend control program is created, allowing acquisition of the simulated queue information of upstream road segments qtjpsjm , the simulated queue information of downstream road segments q. , the simulated queue information of internal road segment q , , the simulated number of vehicles exiting parking lot per unit time V , , and the simulated number of vehicles entering parking lot per unit time V . Additionally, the simulated parking lot entrance and exit lane setup plan is optimized based on the training of the parking lot entrance and exit lane setup model. In the simulation platform, the simulated queue information of the upstream road segments q sim, the simulated queue information of downstream road segments , the simulated queue information of the internal road segments q , , the simulated number of vehicles exiting parking lot per unit time V , ., and the simulated number of vehicles entering parking lot per unit time V are used as the state. The minimum number of exit lanes c . and the minimum number of entrance lanes cm, along with the rules for setting the entrance and exit lanes, are used as constraint conditions. Then, the average queue length of road segments is calculated by an average simulated queue length of upstream road segments, downstream road segments, and internal road segments. The vehicle traffic efficiency is calculated by the simulated number of vehicles exiting parking lot per unit time, and the simulated number of vehicles entering parking lot per unit time. Based on deep reinforcement learning algorithms (including Deep Q-Network, Deep Deterministic Policy Gradient Network, Multi-Agent Deep Deterministic Policy Gradient Network, etc ), the parking lot entrance and exit lane setup model Af0 is constructed with the minimum number of entrance lanes and the minimum number of exit lanes, as well as the rules as action constraints. The output is the parking lot entrance and exit lane setup plan a , which is then implemented in the simulation platform at parking lot entrances and exits to obtain a second state x'. A reward r is used to evaluate the setup plan a , and the interaction data (x,a,r,y^ is employed to train the parking lot entrance and exit lane setup models until convergence, resulting in a trained parking lot entrance and exit lane setup model . The rules for setting the entrance and exit lanes require that neither traffics entering the parking lot nor traffics exiting the parking lot intersect with external traffics, and vehicles entering and exiting the parking lot make right turns. The reward is calculated only when the parking lot entrance and exit lane setup model is training and according to an average queue length of road segments, a vehicle traffic efficiency, and whether the rules for setting the entrance and exit lanes are satisfied. The reward r for the parking lot entrance and exit lane setup model is represented by the average queue length of road segments over a certain period, driving efficiency, and whether the setup meets the rules for setting the entrance and exit lanes. If the rules for setting the entrance and exit lanes are satisfied, the reward is set as zero; if not satisfied, the reward is set to a negative value. If the parking lot entrance and exit lane setup plan a satisfies the rules for setting the entrance and exit lanes, then the parking lot entrance and exit lane setup plan a includes all entrances and exits of the parking lot; if the number of entrances lanes is lower than the minimum number of entrance lanes cin or the number of exits lanes is lower than the minimum number of exit lanes c ,, additional exit lanes and / or entrance lanes are randomly opened to ensure the rules for setting the entrance and exit lanes are satisfied and the numbers of entrance and exit lanes meets the respective the minimum number of entrance lanes and the minimum number of exit lanes; otherwise, the entrance and exit lanes are set according to the rules for setting the entrance and exit lanes. e) Apply the parking lot entrance and exit lane setup model On road segments R , vehicle data on road segments is obtained using millimeter-wave radar within the radar's field of view, which may include geographic coordinates , speed information S* , and detection time T*. Simultaneously, cameras capture vehicle video image data within their visual range D* , including license plate numbers N* obtained through vehicle recognition and feature extraction techniques, external characteristics A^ (such as vehicle type and color), positioning coordinates C^,, detection time T*, and lane information . Based on the positioning coordinates from both the millimeter-wave radar and the cameras, the coordinates of the vehicle v are converted into specific lanes and precise positions within those lanes. Ulis information, including license plate numbers N*, external characteristics , lane information JI, coordinates Cd, speed S*, and detection time T*. is then integrated and recorded. Further, based on vehicle speed, lane information, and detection time, the real-time queue information of upstream road segments q ,. the real-time queue information of downstream road segments q^ ,. and the real-time queue information of internal road segments q , , are extracted. At the same time, using the real-time vehicle data D real collected by cameras at parking lot entrance and exit barrier gates, the number of vehicles exiting parking lot in the past hour and the number of vehicle entering parking lot in the past hour is analyzed to obtain the real-time number of vehicles exiting parking lot per unit time Vmjt real and the real-time number of vehicles entering parking lot per unit time V. in,real After the training of the parking lot entrance and exit lane setup model is completed, real-time data is input into the trained model including the real-time queue information of upstream road segments q ,, the real-time queue information of downstream road segments q. ,, the real-time queue information of internal road segments q . ,, the real-time number of vehicles exiting parking lot per unit time V ,. and the real-time number of vehicles entering parking lot per unit time 1’ ,. The model then outputs the real-time parking lot entrance and exit lane setup plan areaI. The entrances and 8 exits of the parking lot has at least one lane, and the parking lot entrance and exit lane setup plan involves setting lanes as exit lanes or entrance lanes, or closing lanes. f) Guide vehicles within and / or outside the parking lot (optional) When the parking lot entrance and exit lane setup plan changes, if vehicles are inside the parking lot, they can obtain the updated parking lot exit lane setups through the vehicle-to-infrastructure communication. Uns information can be loaded into their high-precision navigation applications, allowing for the replanning of driving paths in mobile navigation applications or an on-board navigation applications. Additionally, electronic display screens within the parking lot provide real-time guidance on vehicle driving directions, helping drivers find the parking lot exit. On the other hand, if drivers are outside the parking lot and need to enter the parking lot after the parking lot entrance and exit lane setup plan has updated, drivers can receive updates on the parking lot entrance and exit lane setups and available parking spaces from roadside variable message signs. The external road segments have at least one roadside variable message signs, where drivers receive updates of said real-time parking lot entrance and exit lane setup plan and a number of available parking spaces. Drivers can also use communication technologies to connect with roadside variable message signs to obtain the parking lot entrance and exit lane setup plan, which can then be loaded into high-precision map applications for driving route replanning. Drivers will then follow the driving routes displayed on the map application to reach the corresponding parking lot entrance. Compared to existing technologies, the present invention has the following beneficial effects: (1) The adaptive control of variable entrances and exits of the present invention comprehensively considers the queue information of the upstream road segments and downstream road segments of the parking lot, as well as the congestion conditions within the parking lot itself. It enables real-time control of tire parking lot entrance and exit lane, ensuring a reasonable distribution of inflow and outflow traffic. This prevents excessive queuing on internal road segments, upstream road segments, and downstream road segments, improves the spatial distribution of vehicle entrance and exit flow rates, accelerates the speed of vehicles exiting and entering the parking lot, and enhances the overall traffic efficiency in the area. (2) This invention can adaptively adjust based on dynamic traffic information and static traffic information, allowing for a rapid recovery to normal levels when the traffic system is impacted. This effectively expands the flexibility margin available for system adjustments. (3) After the parking lot entrance and exit lane setup is completed, the entrance and exit information can be updated in real-time on electronic display screens within the parking lot, roadside variable message signs, and navigation applications, achieving precise navigation. (4) This invention differentiates between parking lot with sufficient historical vehicle data and those lacking such data when predicting the number of vehicles exiting parking lot per unit time and the number of vehicles entering parking lot per unit time. For parking lot with at least a year of historical vehicle data, machine learning algorithms are used to enhance prediction accuracy. For those with less than a year of historical vehicle data, similar parking lot data can be referenced to provide a simple estimate of the number of vehicles exiting parking lot per unit time and the number of vehicles entering the parking lot per unit time, thereby ensuring the operational efficiency of the system. ( 5 ) The present invention is based on a deep reinforcement learning algorithm. When applying multiagent reinforcement learning algorithms, an actor network can select appropriate actions based on the current state, while a critic network estimates the expected returns of those actions. The combination of the actor and critic networks reduces the training difficulty of the deep reinforcement learning model. A substantial amount of historical vehicle data can assist the agents in identifying control strategies with good generalization capabilities. Furthermore, the concept of ‘centralized training and decentralized execution’ allows each entrance and exit to set its lane setup based on local information, such as queue information of internal road segments, upstream road segments, and downstream road segments. Tliis method enhances coordination among different parking lot entrances and exits, improving the operational efficiency of the entire parking system and surrounding roadways. Figures Figure 1. The flowchart of the calculation process for the minimum number of entrance lanes and the minimum number of exit lanes according to the existing methods. Figure 2. Queuing theory- verification. Figure 3. Tire rules for setting the entrance and exit lanes when connecting a parking lot entrance and exit with a road. Figure 4. The flowchart of this invention. Figure 5. Tire structure of the artificial neural network. Figure 6. The structure of the long short-term memory- (LSTM) network. Figure 7. The structure of the random forest. Figure 8. The rules for entrance and exit lane setup when connecting a parking lot entrance and exit yvith a one-way road. Figure 9. The rules for entrance and exit lane setup when connecting a parking lot entrance and exit with a two-way road. Figure 10. The rules for entrance and exit lane setup when connecting parking lot entrances and exits to two roads at one intersection. Figure 11. The rules for entrance and exit lane setup when connecting parking lot entrances and exits to three roads at two intersections. Figure 12. The rules for entrance and exit lane setup when connecting parking lot entrances and exits to four roads at four intersections. Embodiments The following is a detailed description of the present invention in conjunction with the accompanying drawings and specific embodiments. The preparatory work of the present invention involves the design and deployment of the traffic information collection system and the traffic information dissemination system. The traffic information collection system of the present invention relies on data from two types of sensors: video cameras and millimeter-wave radars, to collect vehicle information. For both external road segments and internal road segments, directional cameras are used, and long-range millimeter-wave radars operating in the 79 GHz frequency band are deployed. To ensure consistency between the world geographic coordinates of the video cameras and millimeter-wave radars, both sensors are installed at the same location on roadside poles outside the parking lot or on columns within the parking lot. The specific installation parameters of these sensors can be adjusted based on the required field of view. When the field of view requirement is 100 - 150 m, the sensor installation height can be set to 6 meters, with a downward tilt angle of 10°, and data is collected at a frequency of 25 Hz. The video camera located at the parking lot entrance and exit barrier gates should be installed about 0.5 m in front of the gate, with a sensor height set to 1.5 meters and a downward tilt angle of 20°. Data collection starts when a vehicle is detected arriving. The traffic information dissemination system in this invention utilizes electronic display screens, roadside variable message signs, and navigation applications to publish the parking lot entrance and exit lane setup plans. Specifically, suspended electronic display screens are installed above the lanes within the parking lot, and wall-mounted electronic display screens are placed on parking lot columns. Roadside variable message signs are set up on external road segments, and navigation applications in vehicles that support vehicle-to-infrastructure communication are synchronized with the parking lot entrance and exit lane setups. Within the parking lot, tire electronic display screens receive path planning information provided by the vehicle’s onboard system. These screens display turning directions at intersections and the distance to the parking lot exit. The roadside variable message signs on external road segments show the real-time parking lot entrance and exit lane setup plan, along with the number of available parking spaces. A mobile navigation application or an on-board navigation application recommends an optimal parking lot entrance based on the driver's destination, using dynamic planning methods to determine the best driving route. It also provides voice prompts to notify drivers of changes to the parking lot entrance and the planned route. The present invention relates to a method for adaptive control of variable entrances and exits in a parking lot. The overall technical route is shown in Figure 4, and the main steps include: 1. Collect historical vehicle data of entrances and exits of a parking lot; 2. Predict a number of vehicles exiting parking lot per unit time and a number of vehicles entering parking lot per unit time; 3. Calculate the minimum number of entrance lanes and the minimum number of exit lanes for the parking lot; 4. Train the parking lot entrance and exit lane setup model; 5. Apply the parking lot entrance and exit lane setup model; Optionally, during the application of the parking lot entrance and exit lane setup model, roadside 12 variable message signs, electronic display screens, and navigation applications can be integrated to direct drivers along newly planned routes. The specific implementation methods are as follows. a) Collect historical vehicle data of entrances and exits of a parking lot At the i -th entrance and exit barrier gates of the parking lot, cameras and vehicle identification and feature extraction technologies are used to capture information of vehicle v, including license plate numbers X. physical characteristics (such as vehicle type and color, etc.) T', and detection time T‘. The detection time T’ includes specific days of the week X' and time periods H‘v, as well as weather conditions W‘, and computes parking fees E‘v. All vehicle data at barrier gates is recorded as Dp . The vehicle information captured by millimeter-wave radar and cameras is categorized into historical vehicle data Dp his and real-time data D. real. Parking fee information E* is calculated based on the time a vehicle enters the parking lot Tvm and the time a vehicle exits the parking lot , as detected by the entrance and exit barrier gates, and applying a parking fee calculation function <p , the parking fee for the vehicle v is calculated as (p^T'^ -T‘m). b) Predict a number of vehicles exiting parking lot per unit time and a number of vehicles entering parking lot per unit time For parking lot with at least one year of historical vehicle data, the historical vehicle data Dpof vehicles exiting and entering the parking lot is utilized. This data of parking lot exist and entrance i is detection information of vehicle v such as license plate numbers A . appearance features (including vehicle type, color, etc.) , detection time T‘, and parking fee information . The detection time T’ are categorized into specific days of the week X' and time periods H[ (peak hours and off-peak hours), along with the weather information JF! at the time of detection. Based on historical vehicle data at all parking lot entrances and exits, the number of vehicles exiting and entering the parking lot during different days of the week and time periods is statistically analyzed. As shown m Figure 5, a neural network is then employed to predict the number of vehicles exiting parking lot per unit time and the number of vehicles entering parking lot per unit time Vin fu. y^=L(T,W,X,E) where out represents the vehicles exiting the parking lot; in denotes vehicles entering the parking lot; fout and fin refers to the prediction function for the number of vehicles exiting and entering the parking lot, respectively; T is the encoding for the prediction period, where 1 corresponds to the morning peak hours, 2 to the evening peak hours, and 3 to off-peak hours; W represents the encoding for weather conditions, and X denotes the encoding for the day in a week, with Monday to Sunday coded from 1 to 7. Finally, E represents the parking fee during the prediction period. Variant Plan A: As shown in Figure 6, a Long Short-Term Memory' (LSTM) network is employed to predict the number of vehicles exiting parking lot per unit time and the number of vehicles entering the parking lot per unit time Vtn . This method utilizes a Recurrent Neural Network (RNN) that leverages long-term historical vehicle data to predict the number of vehicles exiting parking lot per unit time VotiJu and the number of vehicles entering the parking lot per unit time VlnJh- The ‘gate’ mechanism within the LSTM enables the network to add or remove information effectively, ensuring that predictions can be made without encountering issues of gradient vanishing or explosion. Variant Plan B: As shown in Figure 7. a random forest algorithm is employed to predict the number of vehicles exiting parking lot per unit time and the number of vehicles entering the parking lot per unit time This method utilizes the bootstrap resampling method to extract multiple samples from the historical vehicle data Dp his. A decision tree model is constructed for each sample, and the final prediction is achieved by combining the results from multiple decision trees, thereby enabling robust prediction of the number of vehicles exiting parking lot per unit time and the number of vehicles entering the parking lot per unit time. For parking lot with less than one year of historical vehicle data, the number of vehicles exiting parking lot per unit time and the number of vehicles entering the parking lot per unit time are determined by the historical vehicle data of completed parking lots with similar land use type characteristics, using the four-step predicting as a base, and then adjusted by an empirical coefficient. The process involves determining land use types of the area occupied by the current parking lot. uk represents the area for a particular land use type k in the current parking lot p , and 5* =ukPlukf represents the same for a completed parking lot f . If the land similarity index for all types meets a predefined threshold 0.8 <S <1.2, the current parking lot is considered similar to the completed parking lot. Thus, the historical vehicle data from the completed parking lot is used to predict the number of vehicles exiting parking lot per unit time &and the number of vehicles entering the parking lot per unit time fu using machine learning algorithms. These predictions are then multiplied by an empirical coefficient to adjust for potential differences. The empirical coefficient 2 can be calculated based on the similarity of a specific land use type or as the average similarity across all land use types. c) Calculate the minimum number of entrance lanes and the minimum number of exit lanes Based on the predicted traffic demand, specifically the number of vehicles exiting parking lot per unit time and the number of vehicles entering parking lot per unit time Vin ., and considering the raise and lower time of parking lot entrance and exit barrier gates (gate arms) as well as vehicle speed, the traffic capacity of a single entrance or exit lane can be calculated. This refers to the number of vehicles that can enter or exit the parking lot through a single lane in an hour. Using this information, the minimum number of entrance lanes and the minimum number of exit lanes to meet traffic demand can be determined, ensuring that the parking lot entrance and exit lanes are sufficient to handle traffic flow when the lane setup plan changes. Based on the number of vehicles exiting parking lot per unit time Vout fu and the number of vehicles entering the parking lot per unit time VinJu, the minimum number of entrance lanes and the minimum number of exit lanes for the parking lot is calculated as follows: where cout and cjn represent the minimum number of entrance lanes and the minimum number of exit lanes, respectively; The notation [*] indicates rounding up to the nearest integer; Vmt and represent the number of vehicles exiting parking lot per unit time and the number of vehicles entering parking lot per unit time, respectively. Cou1 and Cjn denote the traffic capacity of a single exit lane and entrance lane, respectively’, which can be determined based on the road traffic capacity calculation method: 3600 lOOOv t I ‘o 'o where C represents either Cout or Cjn, t0 is the headway time (s); V is the speed (km / h); Zo is the minimum vehicle spacing (m). The values of t0 . v and Zo can be obtained by averaging the vehicle speed, headway time, and spacing at the parking lot entrance and exit barrier gates. d) Train the parking lot entrance and exit lane setup model In traffic simulation platform such as SUMO and VISSIM, a simulation environment for road segments and parking spaces is established based on the actual road traffic network, the location of parking lot entrances and exits, and the layout of internal road segments. At the same time, Synchro, a signal timing optimization software, is used to determine the optimal signal timing plans for intersections throughout the entire road network, which are then set in the traffic simulation platform. Then, based on the number of vehicles exiting parking lot per unit time V and the number of vehicles entering the parking lot per unit time Vin , random choices are generated for the origins and destinations of drivers using a shortest path method, aiming to minimize spatial path lengths. Specifically, the starting point for parking vehicles is the parking lot, while the destination is a location outside of the parking lot. Conversely, the starting point for incoming vehicles is a location outside the parking lot, with the parking lot serving as their destination. An interactive logic and interface are established between the simulation platform and the backend control program to acquire and control the state of road traffic in the traffic simulation environment. During the traffic simulation process, the backend control program continuously retrieves simulated information regarding the simulated queue information of upstream road segments, the simulated queue information of downstream road segments, and the simulated queue information of internal road segments. The traffic simulation platform utilizes the data collected on both external road segments and internal road segments to optimize the parking lot entrance and exit lane setups in real-time using the parking lot entrance and exit lane setup model. This is achieved through the interface between the simulation platform and the backend control program, allowing for effective control of parking lot entrance and exit lanes. Variant Scheme A: Based on swarm intelligence algorithms (such as ant colony algorithm, genetic algorithm, simulated annealing algorithm, etc.), the shortest spatial path is used as the initial solution, with the goal of minimizing travel time. The algorithm iteratively refines the solution to obtain the optimal driving path. In the traffic simulation platform, the simulated queue information of upstream road segments q sim, the simulated queue information of downstream road segments q, , the simulated queue information of internal road segments q . , the simulated number of vehicles exiting parking lot per unit time Vaitsm, and the simulated number of vehicles entering the parking lot per unit time V are set as the state of the parking lot entrance and exit. Constraints are established based on the minimum number of entrance lanes, the minimum number of exit lanes, and the rules for setting the entrance and exit lanes. A deep reinforcement learning algorithm is then employed to construct a parking lot entrance and exit lane setup model. To ensure that the setup of parking lot entrance and exit lanes meets practical application needs, using a common double-lane setup as an example, it is considered that all lane setups are flexible in adaptive entrance and exit lane setups. This means they can be closed or opened, and when opened, lanes can be freely designated as either exit lanes or entrance lanes. Each parking lot entrance and exit works as an independent agent that can determine and change its entrance and exit lanes based on its current state. For instance, taking a specific parking lot entrance and exit i as an example, the state of the entrance and exit be represented as follows: $, _ r .y q i \-±up,sim', cldown,sim^ ±park,sim^ out,sim^ F in,sim J a,(0 = {0,1,2,3} where x, represents the observed state of the parking lot entrance and exit i. q1 indicates the queue information of the upstream road segment of the parking lot entrance and exit i, q'dm,„ sim represents the queue information of the downstream road segment of the parking lot entrance and exit i, and at denotes the parking lot entrance and exit lane setup plan of the parking lot entrance and exit i. The action values are defined as follows. The number 0 indicates that the entrance and exit are closed. The number 1 signifies that the entrance and exit are open with both lanes designated as entrance lanes. The number 2 indicates that the entrance and exit are open with both lanes designated as exit lanes. The number 3 means that the entrance and exit are open with one lane serving as an entrance lane and the other as an exit lane. The parking lot entrance and exit lane setup plan generated by the agent must be implemented in accordance with the rules for setting the entrance and exit lanes. If the setup plan meets these rules, the number of entrance lanes and the number of exit lanes are calculated. If the number of entrances lanes is lower than the minimum number of entrance lanes c or the number of exits lanes is lower than the in minimum number of exit lanes cmt, additional exit lanes and / or entrance lanes are randomly opened to ensure the rules for setting the entrance and exit lanes are satisfied and the numbers of entrance and exit lanes meets the respective the minimum number of entrance lanes and the minimum number of exit lanes; otherwise, the entrance and exit lanes are set according to the rules for setting the entrance and exit lanes. The rules for setting the entrance and exit lanes can be designed based on the connection conditions between the parking lot entrances and exits and the surrounding roadways. Specific case examples are shown in Figure 8 ~ 13. For a parking lot, coordinating the setup of multiple entrances and exits is essential to effectively address congestion caused by uneven distribution of traffic flow in space. The overall state of the parking lot entrances and exits is represented as x =[s15...,sv], and the overall lane setup plan is represented as a = [ax,..., aN ]. For the selected parking lot entrance and exit lane setup plan under this state, during the subsequent evaluation period, which corresponds to the time interval A / , the system continuously evolves using the built-in car following models, lane changing models, and real-time path planning models. After a time interval At, multi-agent deep reinforcement learning evaluates the effectiveness of the parking lot entrance and exit lane setup plan through a reward function. Ihc reward function is utilized to evaluate the impact of the selected parking lot entrance and exit lane setup plan on both dynamic traffic information and static traffic information. Considering the interdependent relationship between the various entrances and exits of the parking lot, the goal of the parking lot entrance and exit lane setup plan is to ensure that the demands for vehicle departures and arrivals are met while achieving coordination among all entrances and exits. This coordination aims to reduce queue lengths of external road segments and internal road segments, thereby improving the efficiency of vehicle departures and arrivals. The overall reward function can be expressed in terms of the average queue lengths on road segments over a specified time period At: where Wj, , if,, if 4, and if 5 are the weights, ql sjm and qdown sim represent the average queue lengths of the upstream road segments and downstream road segments over the time period At, and q rk sim is the average queue length within the parking lot over the time period At. Considering that the entrance and exit lane setup must adhere to the rules for setting the entrance and exit lanes, efforts are made to minimize traffic conflicts. If the generated setup plan a does not satisfy the rules, tire agent gets a penalty, denoted as re = —100. Conversely, if the lane setup a meets the rules, the reward is r = 0. In a word, the reward function is r = rq + re Based on the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) framework, the training of the parking lot entrance and exit lane setup model is established. As the simulation progresses, a substantial amount of interaction data a, r. x is continuously collected from the traffic simulation platform and stored in an experience pool. Following the principle of ‘centralized training’, the global entrance and exit lane setup strategy is optimized using the actions and states of both the current agent and other agents. Depending on the number of entrances and exits in the parking lot, corresponding critic and actor networks are established. A critic network consists of an evaluate network and a target network, and an actor network consists of an evaluate network and a target network. The critic network learns how to evaluate the collaborative the lane setup plan a, while the actor network determines the lane setup plan 17 a for each entrance and exit of the parking lot. Notably, the lane setup plan a for each parking lot entrance and exit is generated by a set of actor networks and evaluated by a set of critic networks. Specifically, each actor network inputs the current state and outputs an action, whereas the critic network outputs the evaluation Q value of the collaborative control effectiveness of parking lot entrances and exits based on the overall state of the parking lot and all actions taken. When the reward function shows a converging trend during the simulation, the model is deemed trained, yielding the trained parking lot entrance and exit lane setup model . During the training process, the loss function for the critic network is defined as follows: 1 / \ 2 {y3 - )) y3 = r3+y(y,'(x'3a'3,,...,a) , where represents the policy function of the evaluate network in the actor network for parking lot entrance and exit i'; S is the total sample size extracted from the experience pool for model training; ^x3 ,a3 ,r3 ,x'3^ denotes the samples drawn from the experience pool; is the critic network corresponding to the parking lot entrance and exit i and the policy function / / ; a3k is the parking lot entrance and exit lane setup plan output by the evaluate network of the actor network for parking lot entrance and exit k in the / -th sample; N is the total number of entrances and exits in the parking lot; y is the discount factor; oJ is the state of the agent observed by the simulation platform; / / 'k represents the policy function of the target network in die actor network for the parking lot entrance and exit k ; a'3k is the lane setup plan output by the target network in the actor network for the parking lot entrance and exit k in the / -th sampled data. The policy gradient function used for updating the actor network is v<x (yysx frx x) .... 0 \ * J where represents the parameters of the evaluate network in the critic network for the parking lot entrance and exit i. Notably, the identification numbers of the parking lot entrances and exits are unique, with i and k used to distinguish between evaluate networks and target networks. For the parking lot entrance and exit i, the update method for the target network is 0^^+(1-^0^ For a parking lot, the total number of entrances and exits is limited. Therefore, to reduce the complexity of network training, the actor networks and critic networks for different agents can share a same structure. For example, in a parking lot with 4 entrances and exits, where each entrance and exit are designed with two lanes, the actor network has the following neuron network layers: 5-50-30-20-1. The critic network has neuron network layer structure of 15-50-30-20-1. The activation function for the hidden layers of the neural networks is ReLU, and the actor network’s final layer uses the sigmoid function, with the output scaled by 3 to map the results to a predefined action space. The output is then rounded to the nearest integer. Additionally, to avoid redundancy, variables that represent the state, such as the simulated queue information of internal road segments q , . the simulated number of vehicles exiting parking lot per unit time V , and the simulated number of vehicles entering parking lot per unit time V are only inputted once in the neural network. The network parameters are set as shown in the following table. Table 1 The setting of neural network parameters. Parameter name Value Actor learning rate 0.0001 Critic learning rate 0.001 Batch size 2048 Bellman discount factor 0.9 Memory size 20000 The queue length of upstream road segments, the queue length of downstream road segments, the queue length of internal road segment, the number of vehicles exiting parking lot per unit time, and the number of vehicles entering parking lot per unit time are all fed into the parking lot entrance and exit lane setup model. In the cyber-physical traffic system, the parking lot entrance and exit lane setup plans are output, setting the parking lot entrance and exit lanes. In practical applications, this model is implemented using a ‘distributed execution’ method, where each entrance and exit only uses its own actor network to make decisions based on local state information, ensuring fast system responses. Once the lane setup changes are made, the cyber-physical traffic system immediately sends the entrance and exit lane setup plan to vehicles and traffic information dissemination system on the road, notifying drivers of the changes at the parking lot entrance and exit lanes. Traffic information dissemination system on the road include electronic display screens inside the parking lot and roadside variable message signs. In addition to the overall parking lot entrance and exit lane setups, the internal electronic display screens also guide vehicles on driving directions within the parking lot. Variant Scheme A: Based on the Deep Q-Network (DQN), the parking lot entrance and exit lane setup model is trained. As the simulation progresses, a large amount of real-time interaction data (x,a,r,x'} from the traffic simulation platform is collected and stored in an experience pool. Using the actions and states of both the current agent and other agents, the lane setup strategy for each parking lot entrance and exit lane is optimized. The DQN outputs the parking lot entrance and exit lane setup plan for all parking lot entrances and exits based on the overall state of the parking lot. Specifically, the chosen lane setup plan for the entrance and exit lanes is the one that maximizes the expected reward, i.e., the optimal policy that leads to the highest cumulative reward: Q*(x,a) = E^s r + ymaxQ*(x',a'}\x,a a' a = maxa Q(s,a;0) where Q * (x, a) is the optimal Q function; £ is the environment, / is the discount factor. In practice, a neural network O^x,a,0^ is used to approximate Q*(x,d), where 0 represents the parameters of the neural network. During network training, the total number of samples for model training is randomly selected from the experience pool. ^X ,u .r ,x' represents a sample drawn from the experience pool, and 5 represent the total number of samples drawn for model training. The loss function of the Q network can then be written as: L = ^ky3 ~Q^3 y = r + y maxa Q (V , ci ; 0)] For a parking lot with a limited number of entrances and exits, the action space of the network is not very large, allowing single-agent reinforcement learning to solve a parking lot entrance and exit lane setup problem. For example, in a parking lot with 4 entrances and exits, each set as a double lane, the Q network would have the following structure with neuron counts in layers: 11-100-50-16, where the activation function for the hidden layers is ReLU. Notably, certain variables that represent the overall state of the parking lot, such as the simulated queue length of internal road segments q , , the simulated number of vehicles exiting parking lot per unit time V , ., and the number of vehicles entering parking lot per unit time V. , are input only once into the neural network to avoid redundancy. The network parameter settings are summarized in the table below: Table 2 The neural network parameter settings. Parameter name Value Actor learning rate 0.0001 Critic learning rate 0.001 Batch size 1024 Bellman discount factor 0.9 Memory size 20000 Variant Scheme B: The parking lot entrance and exit lane setup model is trained based on the Deep Deterministic Policy Gradient (DDPG) network. As the simulation continues, a large amount of interaction data {x, a, r, x') is collected in real-time from the traffic simulation platform and stored in an experience pool. The traffic information physical system is treated as an agent that utilizes the current actions and states to optimize the global entrance and exit lane setup strategy. Critic network and actor network are established, with each network composed of an evaluate network and a target network. The critic network is used to learn how to evaluate the parking lot entrance and exit lane setup strategy, while the actor network determines the setting for each entrance and exit of the parking lot. Notably, the setting for all parking lot entrances and exits are controlled by a set of actor networks and evaluated by a set of critic networks. Specifically, the actor network inputs the states of all parking lot entrances and exits and outputs actions, while the critic network evaluates the current cooperative control effect of the entrance and exit lane setup plan based on the overall state of the parking lot and all actions using the Q value. When the rewards show a trend of convergence during the simulation, the model training is completed, yielding the trained model . The loss function for the critic network during training is given by: where 5 represents the total sample size drawn from the experience pool for model training; r. is the reward; Q'[xt,p.'[xt |0"' )|^' ) denotes the target network in the critic network; represents the evaluate network in the critic network; / / '(xj#" j is the target network in the actor network; 0l! refers to the parameters of the target network in the actor network; dQ and 0Q are the parameters of the target and evaluate networks in the critic network, respectively. The policy gradient function used for updating the actor network is given by: where 0" represents the parameters of tire evaluate network in the actor network for the parking lot entrance and exit lane setup model. For the parking lot entrance and exit, the update method for the target network is as follows: 0' / - T0 + (\-t)0' For a parking lot, the total number of entrances and exits is limited. Therefore, to reduce the difficulty of network training, different agents can use the same structure fortheir actor networks and critic networks. For example, in a parking lot with 4 entrances and exits, where each entrance and exit are designed with two lanes, the actor network has the following neuron network layers: 11-50-30-20-4, and the critic network has the following neuron network layers: 15-50-30-20-1. Hie activation function used for the intermediate layers of the neural netw orks is ReLU. The output layer of the actor network employs the sigmoid function and multiplies the output result by 3 to map it to the predefined action space, while the integer part of the output result is taken. Specifically, in each parking lot entrance and exit state, the variables representing the overall state of the parking lot, such as the queue information of internal road segments q . the simulated number of vehicles exiting parking lot per unit time v , and the simulated number of vehicles entering parking lot per unit time Vm , are input into the neural network only once to avoid duplication. The network parameter settings are shown in the following table: Table 3 The neural network parameter settings. Parameter name Value Actor learning rate 0.0001 Critic learning rate 0.001 Batch size 1024 Bellman discount factor 0.9 Memory size 20000 e) Apply the Parking Lot Entrance and Exit Lane Setup Model On the road segment R , millimeter-wave radar is used to collect vehicle data on road segments DR within its detection range, which may include the vehicle's geographic coordinates CR, speed information SR,. and detection time T*. At the same time, cameras are utilized to capture video image data DR of vehicles within their visual range. This includes vehicle license plate numbers appearance features (such as vehicle type, color, etc.) AR, positioning coordinates C*, detection time T*, and tire lane in which the vehicle is located it . based on vehicle target recognition and feature extraction techniques. Using the positioning coordinates from both the millimeter-wave radar and the cameras, the coordinates of vehicle v are converted into specific lanes and specific positions on those lanes. The vehicle information (including license plate number NR, appearance features AR, lane location LR. coordinates C“, speed S*) and detection time T* are integrated and recorded together. Based on vehicle speed, lane location, and detection time, further the queue lengths of upstream road segments the queue lengths of downstream road segments q, ,, and queue lengths of internal road segments q , , are extracted. Simultaneously, real-time vehicle data collected from the cameras at the parking lot entrance barrier gates D real are used to determine the number of vehicles entering the parking lot in the past hour and the number of vehicles exiting the parking lot in the past hour, allowing for the acquisition of real-time data on the number of vehicles exiting parking lot per unit time V , and the number of vehicles entering the parking lot per unit time Vjn reaI. Real-time road traffic information is collected using two types of sensors: video cameras and millimeterwave radar. On the road segments inside or outside the parking lot R , millimeter-wave radar is used to gather vehicle data on road segments within the radar's field of view DR . which may include global geographic coordinates , speed information SR , and detection time TR . Simultaneously, video cameras capture vehicle video image data within the camera's field of view DI, including license plate numbers N*. vehicle appearance characteristics (such as model, color, etc.) AR. location coordinates C'l. detection time TR, and lane information / 1 , obtained through vehicle target recognition and feature extraction technologies. Based on the positioning coordinates from both the millimeter-wave radar and cameras, the vehicle's positioning coordinates are converted into specific lanes and positions within the lanes. The information about the vehicle v , including speed SR , lane L* , coordinates CR , license plate number N* , appearance characteristics (including model, color, etc.) AR, and detection time T*, is then integrated and recorded. Hie data format is shown in Table 4. Through data fusion technology, the traffic data from various road segments is consolidated to obtain real-time information on the queue lengths of upstream road segments q ,, the queue lengths of downstream road segments q. ,, and the queue lengths of internal road segments q , ,. ® ® 1 park,real Table 4 Real-time road traffic information on internal road segments and external road segments ID License plate number Speed Road segment Lane X (m) Y (m) Vehicle type Color Detected time 1 AXXXX 32 XX Road 01 1.75 5.0 Passenger car Black 10:01 2 iri b xxxx 21 XX Road 02 5.25 1.0 Passenger car Red 10:01 3 iri axxxx 23 XX Road 02 5.25 13.1 Passenger car Blue 10:02 ...... ...... ...... ...... ...... ...... ...... ...... ...... 51 fy D XXXX 18 XX Road 01 1.75 16.4 Passenger car White 10:16 52 fy B XXXX 29 XX Road 01 1.75 7.3 Passenger car Red 10:17 53 it1 AXXXX 30 XX Road 02 5.25 5.5 Passenger car Black 10:17 ...... ...... ...... ...... ...... ...... ...... ...... ...... Cameras installed at parking lot entrance and exit barrier gates are used to capture vehicle appearance and license plates, machine learning algorithms such as support vector machines, convolutional neural networks, and generative adversarial networks are employed to identify vehicle features and license plate information. This process obtains the vehicle's license plate number N‘v, appearance characteristics (including model, color, etc.) A'v, detection time fy , and records the detection time T’ (including specific day of the week X‘v and time period H‘v) along with weather information W‘. Parking fee E‘v is then calculated, forming vehicle data at barrier gate D ,. The data format is shown in Table 5. Table 5 Realtime road traffic information at parking lot entrance and exit barrier gates License plate number Speed Vehicle type Color Entry' time Time period Departure time Expense Date Weath er W AXXXX 10 Passenger car Black 10:30 Off-peak - - 2023.1.25 Sunny iri B XXXX 8 Passenger car White 10:35 Off-peak - - 2023.1.25 Sunny it1 AXXXX 7 Passenger car Black 10:41 Off-peak - - 2023.1.25 Sunny iri AXXXX 7 Passenger car Black - Evening peak 17:20 10.0 2023.1.25 Sunny tri AXXXX 6 Passenger car Red - Evening peak 17:29 15.0 2023.1.25 Sunny iri b xxxx 7 Passenger car White - Evening peak 17:37 10.0 2023.1.25 Sunny From the real-time detection data D , collected by cameras at the parking lot entrance and exit barrier gates, the number of vehicles entering the parking lot over the past hour and the number of vehicles exiting the parking lot over the past hour is used as the real-time number of vehicles exiting parking lot per unit time V , , and the real-time number of vehicles entering parking lot per unit time V ,. out.real » r » r m.real In the application of parking lot entrance and exit lane setup model, real-time data detected by millimeter wave radar and cameras is input into the trained model . The inputs include the real-time queue length of upstream road segments q ,, the real-time queue length of downstream road segments q. the real-time queue length of internal road segments q , ,, the real-time number of vehicles exiting parking lot per unit time Vout Teat, and the real-time number of vehicles entering parking lot per unit time Vinreal. Based on these inputs, the model outputs the real-time parking lot entrance and exit lane setup plan areaj. f) Guide vehicles within and / or outside the parking lot (optional) When the parking lot entrance and exit lane setup plan changes, if a vehicle is inside the parking lot, on one hand, the vehicle can communicate in real-time with parking lot facilities through vehicle-to-infrastructure communication, using indoor positioning and communication technologies such as WLAN, RFID, UWB, and Bluetooth. This allows vehicles to obtain the updated exit lane setup of the parking lot, enabling mobile navigation applications or on-board navigation applications to replan the driving route for the shortest travel time. On the other hand, wall-mounted and hanging electronic display screens inside the parking lot provide real-time guidance on the vehicle's driving direction, facilitating real-time navigation within the parking lot. Map applications can load panoramic and localized maps of the parking lot, displaying the planned route and forward direction to help drivers locate the parking lot exit. If the parking lot entrance and exit lane setup plan changes and a vehicle needs to enter the parking lot from outside, the driver can receive updates about the entrance and exit lane setup plan as well as the number of available parking spaces from roadside variable message signs. The driver can also use communication technology to interact with roadside variable message signs to obtain this information, which can then be loaded into a high-precision map application for route replanning. The vehicle will then follow the navigation information displayed on the map to reach the appropriate parking lot entrance. The aforementioned functionalities, when implemented as software functional units and sold or used as independent products, can be stored on a computer-readable storage medium. Based on this understanding, the technical solutions of the present invention, or the contributions made to the existing technology, can essentially be embodied in the form of software products. This computer software product is stored on a storage medium and includes several instructions that enable a computer device (which may be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage mediums include various media that can store program codes, such as USB flash drives, external hard drives, read-only memory, random access memory, magnetic disks, or optical discs.

Claims

1. A method for adaptive control of variable entrances and exits in a parking lot, involving external road segments and internal road segments; said external road segments include upstream road segments and downstream road segments; said internal road segments include road segments where vehicles drive within said parking lot, and road segments where vehicles queue and / or drive at entrances and exits of said parking lot; said method comprises the following steps:a) Collect historical vehicle data Dp b of entrances and exits of said parking lot;b) Based on said historical vehicle data Dp his, predict a number of vehicles exiting parking lot per unit time and a number of vehicles entering parking lot per unit time Vin ;c) Based on I ’ . and v. . , calculate a minimum number of entrance lanes c. and a minimum number of exit lanes cout for said parking lot;d) In a simulation platform, using simulated queue information of upstream road segments q sjm, simulated queue information of downstream road segments q. , simulated queue information of internal road segments q , , a simulated number of vehicles exiting parking lot per unit time V , and a simulated number of vehicles entering parking lot per unit time V as a state x ; applying constrains of said minimum number of entrance lanes cjn, said minimum number of exit lanes cout, and rules for setting the entrance and exit lanes, establish a parking lot entrance and exit lane setup model Mo, output a parking lot entrance and exit lane setup plan a ; said plan a is then implemented at entrances and exits in said simulation platform to obtain a second state x' • said plan a is evaluated using a reward r, and said parking lot entrance and exit lane setup model M: is trained to convergence using interaction data (x,a,r,x'), ultimately obtaining a trained parking lot entrance and exit lane setup model ;e) Apply real-time queue information of upstream road segments q real-time queue information of downstream road segments q. ,, real-time queue information of internal road segments q , ,, areal-time number of vehicles exiting parking lot per unit time v ,, and a real-time number of vehicles entering parking lot per unit time Vjn real to said model , to obtain a realtime parking lot entrance and exit lane setup plan .

2. A method according to Claim 1, characterized in that, if said parking lot has at least one year of historical vehicle data, said number of vehicles exiting parking lot per unit time VM and said number of vehicles entering parking lot per unit time Vin are calculated from said historical vehicle data using machine learning algorithms; if said parking lot has less than one year of historical vehicle data, predictions are made based on a four-step predicting, using a land similarity 5 to evaluate the degree of similarity between the characteristics of different parking lots; subsequently, the historical vehicle data of another parking lot with similar characteristics, and an empirical coefficient x are applied to predict said number of vehicles exiting parking lot per unit time VM and said number of vehicles entering parking lot per unit time Vjn &3. A method according to Claim 1, characterized in that, said parking lot entrance and exit lane setup plan a is adjusted according to said rules for setting the entrance and exit lanes: if said parking lot entrance and exit lane setup plan a satisfies said rules for setting the entrance and exit lanes, then said parking lot entrance and exit lane setup plan a includes all entrances and exits of said parking lot; if tire number of entrances lanes is lower than said minimum number of entrance lanes cm or the number of exits lanes is lower than said minimum number of exit lanes c ,, additional exit lanes and / or entrance lanes are randomly opened to ensure said rules for setting the entrance and exit lanes are satisfied and the numbers of entrance and exit lanes meets the respective said minimum number of entrance lanes and said minimum number of exit lanes; otherwise, the entrance and exit lanes are set according to said rules for setting the entrance and exit lanes.

4. A method according to Claim 1. characterized in that, said rules for setting the entrance and exit lanes require that neither traffics entering said parking lot nor traffics exiting said parking lot intersect with external traffics, and vehicles entering and exiting said parking lot make right turns.

5. A method according to Claim 1, characterized in that, said reward r is calculated only when said parking lot entrance and exit lane setup model is trained and according to an average queue length of road segments, a vehicle traffic efficiency, and whether said rules for setting the entrance and exit lanes are satisfied; specifically, said average queue length of road segments is calculated by an average queue length of upstream road segments q . an average queue length of downstream road segments a, , and an average queue length of internal road segments q , , all obtained from saidsimulation platform; said vehicle traffic efficiency is calculated by said number of vehicles exiting parking lot per unit time sim and said number of vehicles entering the parking lot per unit time K ; if said rules for setting the entrance and exit lanes are satisfied, said reward is set to ‘O’; if not satisfied, said reward is set to a negative value.

6. A method according to Claim 1, characterized in that, said parking lot entrance and exit lane setup model Mo is trained as follows: based on a deep reinforcement learning algorithm, and a neural network established, models [M0,M0',...] are trained iteratively using said interaction data (x,a,r,x') and said parking lot entrance and exit lane setup model M:, until a converged model, a trained parking lot entrance and exit lane setup model is obtained.

7. A method according to Claim 1, characterized in that, said trained parking lot entrance and exit lane setup model is applied as follows: when said trained parking lot entrance and exit lane setup model is applied to said parking lot, real-time data including said real-time queue information of upstream road segments q ,, said real-time queue information of downstream road segments q^ ,. said real-time queue information of internal road segments q , ,, said real-time number of vehicles exiting parking lot per unit time v „ and said real-time number of vehicles entering parking lot per unit time tz , are used as a state of said trained parking lot entrance and exit lane setup model; Said model Mx then outputs said real-time parking lot entrance and exit lane setup plan areci toadjust entrances and exits of said parking lot.

8. A method according to Claim 1, characterized in that, entrances and exits of said parking lot has at least one lane, and said parking lot entrance and exit lane setup plan involves setting lanes as exit lanes or entrance lanes, or closing lanes.

9. A method according to Claim 1, it further comprises the following optional step:f) After said parking lot entrance and exit lane setup plan is changed, guide vehicles that need to enter said parking lot and vehicles that need to exit said parking lot.

10. A method according to Claim 9, characterized in that, said external road segments have at least one roadside variable message signs, where drivers receive updates of said real-time parking lot entrance and exit lane setup plan and a number of available parking spaces.

11. A method according to Claim 9, characterized in that, said internal road segments include at least one electronic display screen; Based on an updated real-time parking lot entrance and exit lane setup plan areal, said electronic display screens provide real-time direction guidance for drivers, enabling realtime navigation within said parking lot and assisting drivers in locating entrances and exits of said parking lot.

12. A method according to Claim 9, characterized in that, vehicles are equipped with vehicle-to-infrastructure communication and navigation applications; Said navigation applications includes at least one of the following: a mobile navigation application or an on-board navigation application; Said navigation applications can obtain said real-time parking lot entrance and exit lane setup plan areal via vehicle-to-infrastructure communication and load it into high-precision map applications, enabling driving route replanning; Consequently, drivers can follow said driving route provided on said high-precision map applications to reach entrances and exits of said parking lot.

13. A device for adaptive control of variable entrances and exits in said parking lot, comprising a memory, a processor, and a program, characterized in that, said processor implements any of Claims 1-12 when executing said program.

14. A storage medium on which a program is stored, characterized in that, said program is executed to implements any of Claims 1-12.

Images