Automatic driving vehicle cooperation drop-off method and system considering differentiation characteristics

Abstract

The invention discloses an automatic driving vehicle cooperation drop-off method and system considering differentiation characteristics. The method comprises the steps of obtaining to-be-predicted multi-dimensional data; multi-dimensional data to be predicted are input into a vehicle drop time prediction model, the drop time is obtained, the vehicle drop time prediction model is constructed by adopting a single-layer LSTM structure, and the single-layer LSTM structure is used for capturing a dependency relationship in the time sequence data; according to the state information of a certain vehicle and the state information of other vehicles, obtaining a drop-off place; according to the drop-off time and the drop-off place, the expected drop-off time and the expected winding drop-off time are obtained in combination with the current position of the vehicle; determining a drop-off adjustment strategy adopted by the vehicle according to the expected drop-off time and the expected winding drop-off time; and carrying out drop-off according to the drop-off adjustment strategy, and after the drop-off is completed, determining an evacuation strategy to be adopted by the vehicle. The method is used for improving the differentiated drop-off capability of the drop-off area.

Description

Technical Field

[0001] The present invention belongs to the field of autonomous driving technology, and in particular to a method and system for cooperative passenger drop-off of autonomous driving vehicles taking into account differentiated characteristics. Background Art

[0002] With the rapid growth of the number of passengers at the airport and the increasingly prominent differentiated drop-off needs, the airport is a peak traffic area, especially during the morning and evening peak hours, and the frequent entry and exit of vehicles in the drop-off area often leads to congestion. This congestion not only causes delays in passengers getting off, but also affects the traffic efficiency of subsequent vehicles, further increasing the traffic pressure in the drop-off area and surrounding roads. In addition, the drop-off area often has limited parking space resources and cannot accommodate all vehicles at the same time, resulting in long queues at the entrance. At the same time, due to the lack of flexibility of traditional management methods and insufficient ability to regulate peak traffic, it is difficult to make timely adjustments when encountering changes in traffic flow, resulting in instability of traffic flow and increasing the difficulty of drop-off.

[0003] In response to these problems in airport drop-off areas, combined with the research in recent years, the focus has gradually shifted to intelligent traffic management systems and autonomous driving, in order to improve the traffic efficiency of drop-off areas through cooperative drop-off models. At present, there have been some studies on drop-off area management based on traffic simulation platforms. The main methods include managing the number of vehicles entering the drop-off area through queue generation and control strategies to prevent congestion caused by too many vehicles entering; in addition, information sharing and speed control mechanisms between vehicles have also been applied to such research to ensure that vehicles enter the drop-off area in a more orderly manner, thereby reducing unnecessary parking and waiting; and by extracting features, possible drop-off behaviors can be predicted in advance. These studies have proved that intelligent scheduling and collaborative mechanisms can help improve the drop-off situation in the drop-off area and provide data and theoretical support for further optimization.

[0004] Although existing research has alleviated the congestion problem in the drop-off area to a certain extent, there are still some shortcomings in practical application: (1) The drop-off scenario proposed in the research is not consistent with the drop-off area of ​​domestic airports. In the absence of parking lots, passengers are reluctant to queue for a long time in the drop-off area. This differentiated problem is rarely mentioned, which forces some vehicles that are unwilling to queue to join the queue, increasing the waiting time of the system. (2) Existing queue management strategies are often based on preset rules and lack the ability to dynamically respond to real-time traffic flow. Therefore, it is difficult to achieve optimal scheduling when the traffic suddenly increases or decreases. In addition, the queue generation and priority scheduling methods of vehicles are mostly based on simple fleet structures. There is a lack of in-depth research on vehicle queues in complex traffic environments. For example, vehicles in the same queue may have different target drop-off points; different vehicles may also require different drop-off times. (3) Few studies consider how vehicles should leave after dropping off, that is, how vehicles should dissipate. Whether the previous batch of vehicles can dissipate in time will affect whether the next batch of vehicles can successfully enter the drop-off area to drop off, thus resulting in a chain of extended drop-off time. Summary of the invention

[0005] In order to solve the above technical problems, the present invention proposes a cooperative drop-off method and system for autonomous driving vehicles that take differentiated characteristics into consideration, so as to improve the differentiated drop-off capability of the drop-off area and shorten the drop-off time.

[0006] The present invention provides a cooperative drop-off method for autonomous driving vehicles taking into account differentiated features, including:

[0007] Obtain multi-dimensional data to be predicted;

[0008] Inputting the multi-dimensional data to be predicted into a vehicle drop-off time prediction model to obtain the drop-off time, wherein the vehicle drop-off time prediction model is constructed using a single-layer LSTM structure, and the single-layer LSTM structure is used to capture dependencies in time series data;

[0009] Obtain the drop-off location based on the status information of a certain vehicle and the status information of other vehicles;

[0010] According to the drop-off time and the drop-off location, combined with the current position of the vehicle, an expected drop-off time and an expected circling drop-off time are obtained;

[0011] Determining a drop-off adjustment strategy to be adopted by the vehicle according to the expected drop-off time and the expected circle drop-off time;

[0012] Passengers are dropped off according to the drop-off adjustment strategy, and after the drop-off is completed, an evacuation strategy that the vehicle should adopt is determined.

[0013] Optionally, the multi-dimensional data to be predicted includes: the number of passengers, the number of passengers' luggage, the age of passengers, congestion status and weather conditions.

[0014] Optionally, the vehicle drop-off time prediction model includes:

[0015] t a =f(n p , n s , p a ,s t ,s w )

[0016] Among them, t a is the drop-off time of the vehicle, is the number of passengers, is the number of passengers’ luggage, is the age of the passenger, is the congestion status of the road section where the vehicle is located, The weather conditions at the airport drop-off area.

[0017] Optionally, obtaining the drop-off location based on the status information of a vehicle and the status information of other vehicles includes:

[0018] Through the communication module, the vehicle periodically sends its own status information and receives the status information of other vehicles, and defines the vehicle's own status information and the status information of other vehicles as Hello messages;

[0019] According to the Hello message, the vehicle drops off passengers in a manner that vehicles with similar drop-off points go to the same drop-off point together to obtain the drop-off location.

[0020] Optionally, before obtaining the drop-off location based on the Hello message, the method also includes: encrypting the Hello message to ensure that the identity and status information of the vehicle is only used by authorized vehicles, and authenticating the vehicle identity with a digital certificate to ensure that the received information comes from a trusted source.

[0021] Optionally, determining the drop-off adjustment strategy to be adopted by the vehicle according to the expected drop-off time and the expected circle drop-off time includes:

[0022] If the vehicle does not receive the DT information sent by the vehicles parked in the drop-off area, it will receive the contract CONT information. The vehicle that received the contract CONT information will choose whether to accept the speed recommended by the contract CONT information to queue up for drop-off and obtain the selection result;

[0023] According to the selection result, the drop-off adjustment strategy adopted by the vehicle is determined in combination with the expected drop-off time and the expected circle drop-off time.

[0024] Optionally, according to the selection result, in combination with the expected drop-off time and the expected circling drop-off time, determine the drop-off adjustment strategy adopted by the vehicle;

[0025] If going to queue for drop-off at the speed recommended by receiving the contract CONT information, then judge the magnitudes of the expected drop-off time and the expected circling drop-off time;

[0026] If the expected circling drop-off time is greater than or equal to the expected drop-off time, then perform the behavior of queuing for drop-off by sending a queuing instruction to the autonomous vehicle terminal;

[0027] If the expected drop-off time is greater than the expected circling drop-off time, then send an instruction to the autonomous vehicle terminal to adopt the method of having the vehicles with even sequence IDs circle for drop-off and the vehicles with odd sequence IDs queue for drop-off to implement the drop-off behavior.

[0028] Optionally, perform drop-off according to the drop-off adjustment strategy. After complete drop-off, determine the evacuation strategy that the vehicle should adopt, including:

[0029] Through the communication module between vehicles after drop-off, obtain the status information of the vehicles in front of and behind in the driving lane;

[0030] According to the status information, obtain the time headway and adjusted time headway;

[0031] Take the larger value of the time headway and the adjusted time headway as the RDL. According to the RDL and the safe departure time headway SDL, determine the evacuation strategy that the vehicle should adopt.

[0032] Optionally, take the larger value of the time headway and the adjusted time headway as the RDL. Determine the evacuation strategy that the vehicle should adopt, including:

[0033] If RDL SDL, the vehicle in the drop-off area takes a lane-changing behavior to drive away from the drop-off area;

[0034] If RDL < SDL, and RDL 0.6SDL, the vehicles in front of and behind in the driving lane take the behavior of the following vehicle decelerating. The leading vehicle can take the behavior of accelerating if conditions permit, and cooperate to leave a time headway of SDL to help the vehicle in the drop-off area drive away;

[0035] If RDL < SDL, and the waiting time of the vehicle since the end of drop-off has exceeded the maximum waiting time WTM, the vehicles in front of and behind in the driving lane take the behavior of the following vehicle decelerating. The leading vehicle can take the behavior of accelerating if conditions permit, and cooperate to leave a time headway of SDL to help the vehicle in the drop-off area drive away;

[0036] If the vehicle only meets the condition of RDL < SDL, it can only change lanes and drive away after meeting the departure conditions.

[0037] The present invention also provides a cooperative passenger drop-off system for autonomous driving vehicles that takes into account differentiated features, including: a passenger drop-off time differentiated performance module, a passenger drop-off location preference differentiated module, a vehicle waiting for passenger drop-off selection differentiated module, and a vehicle evacuation module;

[0038] The drop-off time differentiation performance module is used to predict the drop-off time based on the data collected by the vehicle and using the vehicle drop-off time prediction model;

[0039] The drop-off location preference differentiation module is used to obtain the drop-off location based on the status information of a certain vehicle and the status information of other vehicles;

[0040] The selection differentiation module of the vehicle waiting for passenger drop-off is used to obtain the expected passenger drop-off time and the expected circling passenger drop-off time according to the passenger drop-off time and the passenger drop-off location, combined with the current position of the vehicle, and determine the passenger drop-off adjustment strategy adopted by the vehicle according to the expected passenger drop-off time and the expected circling passenger drop-off time;

[0041] The vehicle evacuation module is used to drop off passengers according to the drop-off adjustment strategy, and after the drop-off is completed, determine the evacuation strategy that the vehicle should adopt.

[0042] Compared with the prior art, the present invention has the following advantages and technical effects:

[0043] 1. The present invention comprehensively considers the impact of the number of luggage, the number of passengers, the age of passengers, the congestion status, and the weather status on the drop-off time. Through the multivariate regression method of LSTM fusion attention mechanism, the high-precision and high-granularity prediction of the autonomous driving vehicle is achieved under the premise of considering the differences in the drop-off time, and scientifically guides the airport management department to regulate the supply and demand of transportation.

[0044] 2. The present invention takes into account the differences in passengers' preferences for drop-off locations due to the different distances between the check-in terminal and the ticket gate, and uses a priority location strategy to coordinate vehicles near the drop-off points to drop off passengers at the drop-off point.

[0045] 3. The present invention takes into account the actual situation of the actual drop-off area of ​​domestic airports. By replacing and supplementing the circling drop-off strategy, the drop-off time is shortened, the drop-off efficiency is improved, and the selection differences when dropping off are met.

[0046] 4. The present invention takes into account the problem of orderly dispersal of vehicles after the drop-off task is completed, and proposes a vehicle dispersal plan, so that vehicles in the drop-off area can be dispersed in time, reducing the impact on the drop-off of the next batch of vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0048] Figure 1 A flow chart of a cooperative passenger drop-off method for autonomous driving vehicles taking into account differentiated features according to an embodiment of the present invention;

[0049] Figure 2 It is a schematic diagram of the layout of roadside equipment on roads surrounding an airport according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0051] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0052] By optimizing the queue generation mechanism and dynamic scheduling strategy of the airport drop-off area, giving priority to grouping vehicles at the same drop-off point, and providing differentiated management solutions such as CIRCLE strategy for vehicles that do not want to queue, the traffic order in the drop-off area can be effectively improved and congestion can be reduced. The significance of these strategies is to make the vehicle management of the airport drop-off area more intelligent and flexible, improve the overall traffic efficiency, and reduce the ineffective waiting time of vehicles in the drop-off area, thereby providing passengers with a more efficient and smooth travel experience. At the same time, the application of these intelligent scheduling mechanisms can also provide inspiration for traffic management around the airport and lay the foundation for the development of future smart transportation systems.

[0053] This embodiment proposes a cooperative drop-off method for autonomous driving vehicles that takes into account differentiated features, such as Figure 1 As shown, the specific steps include:

[0054] Obtain multi-dimensional data to be predicted;

[0055] The multi-dimensional data to be predicted is input into the vehicle drop-off time prediction model to obtain the drop-off time. The vehicle drop-off time prediction model is constructed using a single-layer LSTM structure, which is used to capture the dependency relationship in time series data.

[0056] Obtain the drop-off location based on the status information of a certain vehicle and the status information of other vehicles;

[0057] According to the drop-off time and the drop-off location, combined with the current position of the vehicle, the expected drop-off time and the expected circle drop-off time are obtained, wherein the expected circle drop-off time is: the length of the drop-off lane L divided by the free flow speed v. The expected drop-off time is: the time required for the previous vehicle to queue up for drop-off;

[0058] Determine the drop-off adjustment strategy to be adopted by the vehicle based on the expected drop-off time and the expected circle drop-off time;

[0059] Drop off passengers according to the drop-off adjustment strategy. After dropping off passengers, determine the evacuation strategy that the vehicle should adopt.

[0060] Furthermore, the multi-dimensional data to be predicted include: the number of passengers, the number of passengers' luggage, the age of passengers, congestion status and weather conditions.

[0061] Specifically, the number of passengers refers to the total number of all passengers in the vehicle.

[0062] The number of passengers' luggage refers to the total number of luggage carried by all passengers in the vehicle.

[0063] Passenger age refers to the age information of all passengers in the vehicle, which is used to assess the speed at which passengers get off the vehicle and the degree of assistance required.

[0064] Congestion status refers to the current traffic status level in the airport drop-off area, which includes free flow, stable traffic, and congestion.

[0065] Weather conditions refer to the real-time weather conditions in the airport drop-off area, including but not limited to rain, snow, fog and other weather conditions that affect the speed and safety of passengers getting off the bus.

[0066] Furthermore, the vehicle drop-off time prediction model includes:

[0067] t a =f(n p , n s , p a ,s t ,s w )

[0068] Among them, t a is the drop-off time of the vehicle, is the number of passengers, is the number of passengers’ luggage, is the age of the passenger, is the congestion status of the road section where the vehicle is located, The weather conditions at the airport drop-off area.

[0069] Further, according to the status information of a certain vehicle and the status information of other vehicles, obtaining the drop-off location includes:

[0070] Through the communication module, the vehicle periodically sends its own status information and receives the status information of other vehicles, and defines the vehicle's own status information and the status information of other vehicles as Hello messages;

[0071] According to the Hello message, vehicles with similar drop-off points go to the same drop-off point to drop off passengers and obtain the drop-off location.

[0072] Specifically, the Hello message refers to the message format for exchanging information between vehicles, including key information such as vehicle ID, drop-off point, current speed, current location, acceleration, drop-off time, etc.

[0073] Furthermore, before obtaining the drop-off location based on the Hello message, it also includes: encrypting the Hello message to ensure that the vehicle's identity and status information is only used by authorized vehicles, and using a digital certificate to authenticate the vehicle's identity to ensure that the received information comes from a trusted source.

[0074] Specifically, the drop-off location preference differentiation module mainly divides the area based on the distance between the passenger check-in terminal and the ticket gate. The passenger drop-off point can be input through flight information, and the communication between vehicles can coordinate to drop off passengers at different drop-off points. This module is implemented through the following steps:

[0075] Step 1: Each vehicle needs to have a communication module (such as DSRC or C-V2X module) to broadcast its status information regularly and listen to the broadcast information of surrounding vehicles. A standardized information format is defined here, namely the Hello message.

[0076] Step 2: The vehicle sends periodic broadcast messages: Every 100 milliseconds (10 times / second), the vehicle automatically sends a Hello message containing its current location, speed, acceleration, drop-off point, drop-off time, and other information.

[0077] Step 3: Receive messages from surrounding vehicles: The vehicle receives Hello messages from nearby vehicles through the V2V module, decodes the message content, and stores and processes it. The vehicle stores the received status information of other vehicles in the local neighboring vehicle table and updates this information in real time for subsequent calculations and decisions.

[0078] Step 4: To ensure the security of V2V communication, the communication content needs to be encrypted and authenticated to prevent unauthorized vehicles or devices from interfering with data transmission. Message encryption part: Encrypt messages such as Hello to ensure that the vehicle's identity and status information are only available to authorized vehicles. Authentication and access control part: Use digital certificates to authenticate the vehicle's identity to ensure that the received information comes from a trusted source and prevent malicious vehicles from sending false data.

[0079] Step 5: After sending and receiving information, the vehicles can make collaborative decisions based on the information and drop off passengers by going to the same drop-off point together with vehicles at similar drop-off points.

[0080] Furthermore, according to the expected drop-off time and the expected circle drop-off time, the drop-off adjustment strategy adopted by the vehicle is determined to include:

[0081] If the vehicle does not receive the DT information sent by the vehicles parked in the drop-off area, it will receive the contract CONT information. The vehicle that received the contract CONT information will choose whether to accept the speed recommended by the contract CONT information to queue up for drop-off and obtain the selection result;

[0082] According to the selection results, combined with the expected drop-off time and the expected circle drop-off time, the drop-off adjustment strategy adopted by the vehicle is determined.

[0083] Specifically, according to the cooperative drop-off process, if the vehicle does not receive the DT message, it will receive the CONT message. The vehicle that receives the CONT message chooses whether to accept the CONT recommended speed to queue up for drop-off. By calculating the expected drop-off time EDT and the expected circle drop-off time CDT, a vehicle selection model is constructed when the drop-off area is full and needs to wait for drop-off. By comparing the queue drop-off time and the circle drop-off time, when the queue drop-off time is long, the method of using even-numbered sequence ID vehicles to circle and odd-numbered sequence ID vehicles to queue up for drop-off is adopted to achieve the drop-off behavior. This module is implemented through the following steps:

[0084] Step 1: Through real-time communication between the autonomous vehicle and the airport MEC server, the real-time location and speed of the vehicle, as well as the target drop-off location and waiting time for drop-off are collected. At the same time, the expected drop-off time (EDT) and the circle drop-off time (CDT) of the vehicle are calculated.

[0085] Step 2: Based on the collected data, the MEC server compares the vehicle's expected drop-off time with the circle drop-off time and determines the drop-off adjustment strategy that the vehicle should adopt:

[0086] 1. If the circle drop-off time is ≥ the expected drop-off time, a queue instruction is sent to the autonomous driving vehicle terminal to queue up for drop-off.

[0087] 2. If the expected drop-off time is greater than the circle drop-off time, the drop-off behavior is achieved by sending instructions to the autonomous driving vehicle terminal, and the vehicles with even-numbered sequence IDs circle to drop off passengers while the vehicles with odd-numbered sequence IDs queue up to drop off passengers.

[0088] DT information refers to the information received from the vehicles parked in the drop-off area when there are parked vehicles in the drop-off area. DT information contains the expected departure time LV of the cluster head of the vehicle parked in the drop-off area. The vehicle receiving the DT information will estimate its chance of obtaining the space based on its own position and current conditions.

[0089] LV refers to the expected departure time of the cluster head, which is equal to the earliest departure time of the current drop-off area + the expected stay time of the cluster head.

[0090] The CONT information refers to the ID of the vehicle in the queue, the queue ID that distinguishes the queues, namely GID, the recommended speed S, and the expected departure time LV of the cluster head.

[0091] Recommended speed S calculation formula:

[0092]

[0093] in, is the distance between the vehicle and the drop-off area, It is the estimated departure time of the previous cluster head minus the current time, that is, the earliest vehicle expected to leave the drop-off area minus the current time. and are the maximum and minimum speeds under the assumed road rules respectively, and the calculated speed S must be within the speed range of the road.

[0094] The expected drop-off time (EDT) is the time required for passengers to drop off when the drop-off area is full.

[0095] The expected circle drop-off time CDT refers to the time required for the passenger to return to the drop-off area for drop-off after circling once when the drop-off area is full.

[0096] Furthermore, the drop-off is carried out according to the drop-off adjustment strategy. After the drop-off is completed, the evacuation strategy to be adopted by the vehicle is determined to include:

[0097] Through the communication module between vehicles after dropping off passengers, the status information of the vehicles in front and behind the vehicle in the driving lane is obtained;

[0098] According to the status information, obtain the headway and adjust the headway;

[0099] The larger value of the headway and the adjusted headway is taken as RDL. The evacuation strategy that the vehicle should adopt is determined based on RDL and the safe exit headway SDL.

[0100] Specifically, after the drop-off is completed, by comparing the actual headway and the safe headway, supplemented by the maximum waiting time indicator, a clear evaluation indicator is given to cooperate in leaving the drop-off area. This module is implemented through the following steps:

[0101] Step 1: Obtain the current positions, speeds, and acceleration information of the vehicles in front and behind on the driving lane through V2V communication between vehicles, and transmit them to the on-vehicle terminal of the autonomous vehicle.

[0102] Step 2: Calculate the time headway and adjusted time headway based on the obtained information as follows:

[0103]

[0104]

[0105] Where, is the position difference between the front and rear vehicles, and correspond to the speeds of the front and rear vehicles respectively.

[0106] Step 3: Compare the calculated time headway and adjusted time headway, take the larger value as the RDL, and determine the evacuation strategy that the vehicle should adopt:

[0107] 1. If RDL SDL, the vehicle in the drop-off area takes a lane-changing behavior and drives out of the drop-off area.

[0108] 2. If RDL < SDL, and RDL 0.6SDL, the front and rear vehicles on the driving lane take the behavior of the rear vehicle decelerating, and the front vehicle can take the behavior of accelerating if conditions permit, and cooperate to leave a time headway of SDL to help the vehicle in the drop-off area drive away.

[0109] 3. If RDL < SDL, and the waiting time of the vehicle since the end of drop-off has exceeded WTM, the front and rear vehicles on the driving lane take the behavior of the rear vehicle decelerating, and the front vehicle can take the behavior of accelerating if conditions permit, and cooperate to leave a time headway of SDL to help the vehicle in the drop-off area drive away.

[0110] 4. If the vehicle only meets the condition of RDL < SDL, it can only change lanes and drive away after meeting the departure condition, where the departure condition is the above cases 1 - 3.

[0111] RDL refers to the larger value of the calculated time headway and adjusted time headway.

[0112] SDL refers to the safe departure time headway of the vehicle that can safely drive out of the drop-off area.

[0113] WTM refers to the maximum waiting time of the vehicle starting from the end of drop-off and preparing to drive out of the drop-off area.

[0114] This embodiment also provides a cooperative drop-off system for autonomous driving vehicles that takes into account differentiated characteristics, including: a drop-off time differentiation performance module, a drop-off location preference differentiation module, a vehicle waiting for drop-off selection differentiation module, and a vehicle evacuation module;

[0115] The drop-off time differentiation performance module is used to predict the drop-off time based on the data collected by the vehicle and the vehicle drop-off time prediction model;

[0116] A drop-off location preference differentiation module is used to obtain the drop-off location based on the status information of a certain vehicle and the status information of other vehicles;

[0117] The selection differentiation module of the vehicle waiting for passenger drop-off is used to obtain the expected passenger drop-off time and the expected circling passenger drop-off time according to the passenger drop-off time and the passenger drop-off location, combined with the current position of the vehicle, and determine the passenger drop-off adjustment strategy adopted by the vehicle according to the expected passenger drop-off time and the expected circling passenger drop-off time;

[0118] The vehicle evacuation module is used to drop off passengers according to the drop-off adjustment strategy and determine the evacuation strategy that the vehicle should adopt after the drop-off is completed.

[0119] The present embodiment is described in detail below with reference to the accompanying drawings:

[0120] This embodiment proposes a cooperative drop-off method for autonomous driving vehicles that takes into account differentiated features, such as Figure 1 As shown, the specific steps include:

[0121] Step 1: In the drop-off time differentiation performance module, the autonomous driving vehicle needs to be equipped with an integrated sensor system. The system includes but is not limited to the following sensors: passenger detection sensor, passenger baggage quantity sensor and vehicle positioning system. The passenger detection sensor is used to identify the number of passengers in the vehicle and their age distribution in real time, while the passenger baggage quantity sensor is responsible for detecting the number of pieces of luggage in the vehicle. In addition, the roads around the airport will be equipped with traffic flow monitoring systems and meteorological monitoring equipment such as Figure 2 The traffic flow monitoring system uses high-definition cameras and traffic flow sensors to monitor the traffic congestion status of the road in real time, and the meteorological monitoring equipment is used to collect real-time climate data.

[0122] Step 2: The data format of the drop-off time differentiation performance module includes but is not limited to the following: number of passengers (e.g., 3 people), age of passengers (e.g., 28 years old), number of luggage (e.g., 1 piece), congestion status (e.g., free flow), weather conditions (e.g., cloudy and rainy). The frequency of data updates can be adjusted according to actual needs. It is recommended to update the number of passengers, age, and number of luggage in real time every time a passenger gets on or off the bus, and the data for traffic congestion status and weather conditions can be automatically updated every 3-5 minutes.

[0123] Step 3: Data collection is carried out through sensors and monitoring equipment installed on vehicles and roads around the airport. The collected real-time data will be transmitted to the MEC (edge ​​computing) server for processing. First, the MEC server will clean the uploaded data, including outlier detection and missing value supplementation. For example, the data on the number of luggage and the number of passengers will be checked for negative values ​​or abnormally large values ​​that exceed the vehicle's carrying capacity. Such abnormal data can be replaced by the median or mean, or directly removed from the data set. For the age of passengers, the system will verify whether there are negative values ​​or age data that exceeds the actual range. Abnormal data can be filtered by setting an upper limit (such as 80 years old). For traffic congestion status and weather conditions, the system needs to ensure the consistency of all category labels. If there are spelling errors or non-standard labels, they should be merged or removed. For missing data, appropriate filling strategies will be adopted according to the data type. For example, missing values ​​of the number of luggage and the number of passengers can be filled with zero or mean; passenger age can be filled with mean or median; and for categorical variables (such as congestion status and weather conditions), missing values ​​can be filled with the most frequent category. Secondly, the MEC server will standardize the cleaned data and use the Z-score method to standardize continuous variables (such as the number of passengers, the number of luggage, and the age of passengers). The specific calculation formula is shown in Formula (1).

[0124] (1)

[0125] Among them, X is the original data, μ is the mean, and σ is the standard deviation.

[0126] For congestion status and weather status, one-hot encoding is used to assign a binary column to each category, as shown in Table 1 and Table 2. Free flow can be represented as [1, 0, 0], and clear weather can be represented as [1, 0, 0].

[0127] Table 1

[0128] Original state Free Flow Stable traffic Congestion Free Flow 1 0 0 Stable traffic 0 1 0 Congestion 0 0 1

[0129] Table 2

[0130] Original state sunny Rainy Snow, fog sunny 1 0 0 Rainy 0 1 0 Snow, fog 0 0 1

[0131] Step 4: Deploy a vehicle drop-off time prediction model based on the LSTM-Attention method on the MEC server to accurately predict the drop-off time based on multi-source data (number of passengers, number of passengers' luggage, age of passengers, congestion status, and weather conditions).

[0132] (1) Model design:

[0133] LSTM layer configuration: This model uses a single-layer LSTM structure, which contains 128 units and is designed to capture long-term dependencies in time series data. In order to prevent overfitting, a Dropout layer is inserted between LSTM layers, and the Dropout rate is set to 0.3 to improve the generalization ability of the model.

[0134] Attention mechanism: The output of the LSTM layer (i.e., the hidden state) will be used as the input of the Attention layer. By calculating the attention weight of each time step and taking the weighted sum, a vector representing the context is generated to strengthen the model's attention to key information.

[0135] Output layer: The fully connected layer converts the context vector into the final prediction value, which is the vehicle’s drop-off time.

[0136] (2) Model training:

[0137] Optimizer and loss function: During the training process, the Adam optimizer is used to update the model parameters, and the learning rate is set to 0.001 to ensure stable convergence. The mean square error (MSE) is used as the loss function to optimize the accuracy of the drop-off time prediction.

[0138] Batch size and training epochs: Set the batch size to 64 and adjust the number of training epochs based on the size and complexity of the dataset to ensure adequate training.

[0139] (3) Model evaluation and tuning:

[0140] Performance evaluation: The model is evaluated on the test set. The main evaluation indicators include root mean square error (RMSE) and mean absolute error (MAE) to measure the prediction accuracy of the model.

[0141] Parameter tuning: By experimenting with different numbers of LSTM units, learning rates, and Dropout rates, combined with the cross-validation method, we can find the optimal hyperparameter combination to improve the performance of the model.

[0142] Step 5: Inputs of multi-dimensional data including the number of passengers, the number of passengers’ luggage, the age of passengers, traffic congestion status, and weather conditions are passed into the vehicle drop-off time prediction model based on the LSTM-Attention mechanism. The prediction result output by the model is the vehicle drop-off time at the airport drop-off area. The final result will be stored in the vehicle terminal and combined with the differentiated performance module of the drop-off location to perform cooperative drop-off.

[0143] Step 6: After inputting the differentiated drop-off time and drop-off location preference information, vehicles start with periodic broadcast messages (called "Hello" messages). As vehicles broadcast and listen to each other (within range), they can build knowledge about the distances of surrounding vehicles to the destination. Vehicles closer to (but not yet reached) the drop-off area can receive Drop Timing (DT) information, which comes from the Drop Timing zone, that is, information received from parked vehicles in the case of parked vehicles. The DT information contains the departure time (LV) information of vehicles parked in the drop-off area space. Vehicles that receive the DT information will estimate their chances of getting a space based on their own location and current conditions.

[0144] Step 7: The way to estimate the space you get based on your position and current conditions is as follows: if you drive at the current speed and arrive at the drop-off area when at least one vehicle is leaving, you can go to the drop-off area; otherwise, you can adjust your speed so that when you arrive at the drop-off area, at least one parked vehicle is about to leave; then you announce your new speed to the following vehicles by sending a CONT (short for CONTRACT) message, the purpose is to influence the following vehicles to also reduce their speed, and determine the space of the drop-off area according to the order in which the vehicles arrive. The CONT message contains (ID; GID; S; LV) attributes, where GID is the queue ID that distinguishes the queues, S is the recommended speed, and LV is the expected departure time of the cluster head, which is equal to the earliest departure time of the vehicles currently parked in the drop-off area plus the expected stay time of the cluster head. For example, the earliest departure time of the current parking extracted from the DT message is 10:20 am, and the expected stay time of the cluster head is 10 minutes, then LV = 10:30 am. Among all the vehicles that accept the queue, the vehicles that tend to the same drop-off point are prioritized in one row.

[0145] Recommended speed calculation:

[0146] (2)

[0147] in, is the distance between the vehicle and the drop-off area, It is the estimated departure time of the previous cluster head minus the current time, that is, the earliest vehicle expected to leave the drop-off area minus the current time. and are the maximum and minimum speeds under the assumed road rules respectively, and the calculated speed S must be within the speed range of the road.

[0148] Step 8: Selection differentiation module for vehicles waiting for drop-off. Based on the calculated EDT and CDT, if the vehicle receiving the CONT message can accept (and "agree") with the suggested speed and adjust its speed (to the same as the reference vehicle), it announces its acceptance through a member message (MM), which contains the GID and the MID representing the member ID calculated based on the vehicle sequence relative to the reference vehicle, which can be obtained by receiving the "Hello" message. For example, in cooperative drop-off (CDO), the message used when forming a row when approaching the drop-off area. (1) The license plate numbers of the entering vehicles are 1, 2, 3, ... calculated based on the Hello messages sent between vehicles; these vehicles receive the DT information from the first parked vehicle, so they know when the vehicles in the drop-off area leave. (2) Vehicle No. 3 determines that there is no opportunity to drop off in the drop-off area under the current conditions and broadcasts the CONT message (suggesting to form a team). (3) Vehicle IDs 4 and 5 receive the CONT message, accept the agreement to join the queue with vehicle ID 3 as the leader (adjusting their speed relative to the leader), and announce this message through an MM message containing their vehicle order in the queue (MID), which is equal to 2 and 3 respectively. The benefit of accepting the CONT message is that following the speed suggested in the CONT message is to improve the chance of finding space instead of competing.

[0149] The size of the group is set based on the size of the drop-off area:

[0150] A vehicle with a member ID greater than the size of the drop-off area can be represented as the reference vehicle / cluster head of the next row, and its speed is adjusted according to the LV field contained in the CONT message it receives, and so on. Therefore, vehicles entering the drop-off area will be grouped according to their adjusted speed approaching the drop-off area and the size of the drop-off area.

[0151] If the vehicle is unwilling to accept the recommended speed to form a queue: (no stopping strategy) it can choose to go around and come back to drop off passengers.

[0152] Step 9: Any conflicts in the member IDs (MIDs) of the vehicles in the queue will be resolved by our proposed verification mechanism:

[0153] The CDO algorithm also uses a verification mechanism called Group Member Correction mechanism (GMCM) to correct incorrect group member IDs. Vehicles can issue announcements for incorrect member IDs due to misinformation or selfish behavior.

[0154] For example, a vehicle may not receive or be delayed in receiving a CONT or Hello message due to radio traffic congestion, or the vehicle may have just entered the main road from a side road. In the case of selfish behavior, a vehicle can pretend to be part of a convoy closer to the drop-off area rather than a convoy farther away in order to reach the drop-off area faster.

[0155] GMCM is a decentralized approach based on advice messages shared by surrounding vehicles. When at least two vehicles are found to be providing the same advice on the correct vehicle order in a platoon, a modification of the advice message can be accepted that is based on the "Hello" message received from the closer vehicle. This adds a level of reliability to the advice messages as they require two witnesses to be accepted, but does not protect against collusion attacks. There are three identified scenarios associated with failures in the GID and MID.

[0156] Case 1: The error is related to the MID, for example, vehicle ID 5 publishes an MM message with MID equal to 2 instead of 3. Each member of the same group can introduce an Advice Message (AM) that includes the ID of the wrong and correct suggested member, i.e., the MID.

[0157] Case 2: The other two aspects are related to the location of the cluster heads.

[0158] One is that the vehicle announces itself as the cluster head of the queue by sending a CONT message when it is in the member position of the queue queue. Subsequently, after the vehicle accepts the suggestion message, it must broadcast a cluster change (CC) message containing the correct values ​​of GID and MID to inform the subsequent vehicles that it can accept the CONT message to modify its plan. For example, vehicle ID4 broadcasts a CONT message, and vehicle ID5 accepts the agreement; vehicle id2 and vehicle id3 can send a suggestion message to vehicle ID4 to modify the vehicle row position from the cluster head to the GID1 row member with MID2. After vehicle ID4 accepts the suggestion information; it must send a CC message to vehicle ID5, asking it to modify the vehicle position to MID3 in the GID1 row.

[0159] In another case, a vehicle declares itself as a member of the platoon but is actually the cluster head and therefore needs to send a CONT message. In this case, the Proposal message acts as a leader election message which contains the proposed GID, MID and the proposed departure time (LV) of the previous cluster head. Therefore, the elected vehicle must send a CONT message after accepting the leadership role. When the vehicle manages to find a space and stops in the drop-off zone, the vehicle's CDO participates in the stop.

[0160] Step 10: Vehicle dissipation module, after the drop-off is completed, calculates and compares the RDL based on the information broadcast by the vehicles on the driving lane, and presets the safe exit distance SDL and the maximum waiting time;

[0161] Determine the RDL method:

[0162] (3)

[0163] (4)

[0164] Wherein, is the position difference between the front and rear vehicles, and correspond to the speeds of the front and rear vehicles respectively. The larger value among them is determined as the RDL.

[0165] Continue to judge the size of RDL and SDL:

[0166] 1. If RDL SDL, the vehicle in the passenger dropping-off area takes a lane-changing behavior and drives out of the passenger dropping-off area.

[0167] 2. If RDL < SDL, and RDL 0.6SDL, the front and rear vehicles on the driving lane take the behavior of the rear vehicle decelerating, and the front vehicle can take the behavior of accelerating if conditions permit, and cooperate to leave a headway of SDL to help the vehicle in the passenger dropping-off area drive away.

[0168] 3. If RDL < SDL, and the waiting time of the vehicle since the end of passenger dropping-off has exceeded WTM, the front and rear vehicles on the driving lane take the behavior of the rear vehicle decelerating, and the front vehicle can take the behavior of accelerating if conditions permit, and cooperate to leave a headway of SDL to help the vehicle in the passenger dropping-off area drive away.

[0169] 4. If the vehicle only meets the condition of RDL < SDL, it needs to wait until at least one of the above conditions is met before it can change lanes and drive away.

[0170] The above is only a preferred specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the technical field of the present application within the technical scope disclosed by the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A cooperative passenger drop-off method for autonomous driving vehicles considering differentiated features, characterized in that: include: Obtain multi-dimensional data to be predicted; Inputting the multi-dimensional data to be predicted into a vehicle drop-off time prediction model, wherein the vehicle drop-off time prediction model is constructed using a single-layer LSTM structure, and the single-layer LSTM structure is used to capture dependencies in time series data; Obtain the drop-off location based on the status information of a certain vehicle and the status information of other vehicles; According to the drop-off time and the drop-off location, combined with the current position of the vehicle, an expected drop-off time and an expected circling drop-off time are obtained; Determining a drop-off adjustment strategy to be adopted by the vehicle according to the expected drop-off time and the expected circle drop-off time; Passengers are dropped off according to the drop-off adjustment strategy, and after the drop-off is completed, an evacuation strategy that the vehicle should adopt is determined.

2. The autonomous driving vehicle cooperative drop-off method considering differentiated features according to claim 1, characterized in that: The multi-dimensional data to be predicted include: the number of passengers, the number of passengers' luggage, the age of passengers, congestion status and weather conditions.

3. The autonomous driving vehicle cooperative drop-off method considering differentiated features according to claim 2, characterized in that: The vehicle drop-off time prediction model includes: t a =f(n p ,n s ,p a ,s t ,s w ) Among them, t a is the drop-off time of the vehicle, is the number of passengers, is the number of passengers’ luggage, is the age of the passenger, is the congestion status of the road section where the vehicle is located, The weather conditions at the airport drop-off area.

4. The autonomous driving vehicle cooperative drop-off method considering differentiated features according to claim 1, characterized in that: According to the status information of a vehicle and the status information of other vehicles, the drop-off location is obtained including: Through the communication module, the vehicle periodically sends its own status information and receives the status information of other vehicles, and defines the vehicle's own status information and the status information of other vehicles as Hello messages; According to the Hello message, the vehicle drops off passengers in a manner that vehicles with similar drop-off points go to the same drop-off point together to obtain the drop-off location.

5. The autonomous driving vehicle cooperative drop-off method considering differentiated features according to claim 4, characterized in that: Before obtaining the drop-off location according to the Hello message, the method also includes: encrypting the Hello message to ensure that the identity and status information of the vehicle is only used by authorized vehicles, and authenticating the vehicle identity with a digital certificate to ensure that the received information comes from a trusted source.

6. The autonomous driving vehicle cooperative drop-off method considering differentiated features according to claim 1, characterized in that: According to the expected drop-off time and the expected circle drop-off time, determining the drop-off adjustment strategy to be adopted by the vehicle includes: If the vehicle does not receive the DT information sent by the vehicles parked in the drop-off area, it will receive the contract CONT information. The vehicle that received the contract CONT information will choose whether to accept the speed recommended by the contract CONT information to queue up for drop-off and obtain the selection result; According to the selection result, the drop-off adjustment strategy adopted by the vehicle is determined in combination with the expected drop-off time and the expected circle drop-off time.

7. The autonomous driving vehicle cooperative drop-off method considering differentiated features according to claim 6, characterized in that: According to the selection result, combined with the expected drop-off time and the expected circle drop-off time, determine the drop-off adjustment strategy to be adopted by the vehicle; If the speed recommended by the contract CONT information is received, the expected drop-off time and the expected circle drop-off time are determined; If the expected circling drop-off time is greater than or equal to the expected drop-off time, a queuing instruction is sent to the autonomous driving vehicle terminal to perform a queuing drop-off behavior; If the expected drop-off time is greater than the expected circle-drop-off time, the drop-off behavior is achieved by sending instructions to the autonomous driving vehicle terminal, so that vehicles with even-numbered sequence IDs circle to drop off passengers while vehicles with odd-numbered sequence IDs queue up to drop off passengers.

8. The autonomous driving vehicle cooperative drop-off method considering differentiated features according to claim 1, characterized in that: Passengers are dropped off according to the drop-off adjustment strategy. After the drop-off is complete, the evacuation strategy to be adopted by the vehicle includes: Through the communication module between vehicles after dropping off passengers, the status information of the vehicles in front and behind the vehicle in the driving lane is obtained; According to the state information, obtaining the headway and adjusting the headway; Take the larger value between the headway and the adjusted headway as the RDL, and determine the evacuation strategy that the vehicle should adopt according to the RDL and the safe departure headway SDL.

9. The autonomous driving vehicle cooperative drop-off method considering differentiated features according to claim 8, characterized in that: Taking the larger value between the headway and the adjusted headway as the RDL, the determined evacuation strategies that the vehicle should adopt include: If RDL SDL, vehicles in the drop-off area change lanes and leave the drop-off area; If RDL < SDL and RDL < 0.6SDL, the vehicle in front and behind on the driving lane will take the behavior of the following vehicle decelerating. When conditions permit, the leading vehicle can take the behavior of accelerating, and cooperate to leave a headway of SDL to help the vehicle in the drop-off area drive away; If RDL < SDL, and the waiting time of the vehicle since the end of passenger alighting has exceeded the maximum waiting time WTM, the vehicle behind and in front on the driving lane shall perform the behavior of the following vehicle decelerating, and the leading vehicle can perform the accelerating behavior if conditions permit, and cooperate to leave a headway of SDL to help the vehicle in the passenger alighting area drive away; If the vehicle only meets the condition of RDL < SDL, it needs to change lanes and drive away after meeting the vehicle departure conditions.

10. The autonomous driving vehicle cooperative drop-off system considering differentiated features is characterized by: Including: The passenger alighting time differentiation performance module, the passenger alighting location preference differentiation module, the vehicle waiting for passenger alighting selection differentiation module, and the vehicle evacuation module; The passenger alighting time differentiation performance module is used to predict the passenger alighting time according to the data collected by the vehicle by using the vehicle passenger alighting time prediction model; The passenger alighting location preference differentiation module is used to obtain the passenger alighting location according to the status information of a certain vehicle and the status information of other vehicles; The vehicle waiting for passenger alighting selection differentiation module is used to obtain the expected passenger alighting time and the expected loop passenger alighting time according to the passenger alighting time and the passenger alighting location, combined with the current position of the vehicle, and determine the passenger alighting adjustment strategy adopted by the vehicle according to the expected passenger alighting time and the expected loop passenger alighting time; The vehicle evacuation module is used to perform passenger alighting according to the passenger alighting adjustment strategy, and after completing passenger alighting, determine the evacuation strategy that the vehicle should adopt.

Images