A method for controlling intelligent networked vehicles in a human-machine mixed driving traffic flow in an expressway interweaving area

Abstract

The application relates to a fast road interweaving area man-machine mixed driving traffic flow intelligent network connected vehicle control method, and belongs to the technical field of intelligent transportation, wherein the method comprises the following steps: acquiring historical state data of a target vehicle and surrounding vehicles of the target vehicle; inputting the historical state data into a trained lane-changing intention prediction model to obtain a lane-changing intention of the target vehicle; the trained lane-changing intention prediction model comprises an input layer, a CNN layer, an LSTM layer, an Attention layer and an output layer connected in sequence; when the lane-changing intention is lane changing, a cooperative vehicle on the target lane is controlled to execute a cooperative lane-changing strategy. The application constructs a lane-changing intention prediction model of a fast road interweaving area based on a CNN, an LSTM and an attention mechanism for a complex traffic scene of the interweaving area, effectively and accurately predicts the lane-changing intention of an intelligent network connected vehicle in the fast road interweaving area, and controls the vehicle to execute a corresponding lane-changing strategy.

Description

Technical Field

[0001] This application relates to the field of intelligent transportation technology, and in particular to an intelligent connected vehicle control method for human-machine mixed driving traffic flow in expressway weaving areas. Background Technology

[0002] In recent years, with the rapid development of intelligent transportation systems and intelligent driving technologies, intelligent driving vehicles have gradually penetrated urban open roads. It can be said that intelligent connected vehicles will be important participants and rule changers in future urban road traffic. Although the widespread adoption of intelligent connected vehicles cannot be achieved in the short term, in the foreseeable future, there will be a long-term situation in urban road traffic systems where autonomous vehicles and human-driven vehicles coexist, forming a complex mixed traffic environment.

[0003] The weaving areas of expressways, as nodes where main roads meet auxiliary roads or ramps, play a decisive role in the overall traffic efficiency of the expressway system. Because drivers have extremely limited time and space to maneuver when merging in these weaving areas, they are often considered congestion points in urban expressway traffic. Compared to other parts of the expressway, the weaving areas have high vehicle density, and frequent deceleration, acceleration, and lane changes make the traffic conditions extremely complex, easily leading to traffic accidents and affecting traffic efficiency. Therefore, accurate and effective prediction of the lane-changing intentions of intelligent connected vehicles is crucial for enabling them to make corresponding decisions, plans, and controls in advance, reducing traffic conflicts and improving traffic efficiency and safety.

[0004] In existing technologies, the selection of input features for predicting lane change intentions often uses the position, speed, and acceleration information of the target vehicle as model inputs, failing to integrate and consider the absolute and relative motion information of surrounding vehicles. This is especially true for complex traffic scenarios in weaving areas, where it is difficult to fully learn vehicle driving behavior and its interaction relationships, resulting in poor prediction accuracy. Consequently, this affects the behavioral decisions of intelligent connected vehicles, impacting the traffic efficiency and safety of expressway weaving areas. Summary of the Invention

[0005] In view of this, it is necessary to provide an intelligent connected vehicle control method for human-machine mixed driving traffic flow in expressway weaving areas, in order to solve the problem that the existing technology is difficult to fully learn the vehicle driving behavior and its interaction relationship, resulting in poor prediction accuracy and affecting the traffic efficiency and safety of expressway weaving areas.

[0006] To address the aforementioned issues, this application provides an intelligent connected vehicle control method for human-machine mixed-driving traffic flow in expressway weaving zones, comprising:

[0007] Acquire historical status data of the target vehicle and its surrounding vehicles;

[0008] The historical state data is input into a fully trained lane change intention prediction model to obtain the lane change intention of the target vehicle. The fully trained lane change intention prediction model includes an input layer, a CNN layer, an LSTM layer, an Attention layer, and an output layer connected in sequence.

[0009] When the lane-changing intention is to change lanes, the target vehicle and cooperating vehicles in the target lane are controlled to execute a coordinated lane-changing strategy.

[0010] In some possible implementations, the fully trained lane change intention prediction model is obtained through the following steps:

[0011] Obtain the CitySim dataset and extract the driving trajectory of each vehicle from the CitySim dataset;

[0012] Extract lane-keeping trajectories from the driving trajectory where the lane does not change, take the last moment of the lane-keeping trajectory as the first end moment, take the moment before the first end moment as the first start moment, and take the time from the first start moment to the first end moment as a lane-keeping scenario.

[0013] The moment when the vehicle crosses the lane line is extracted from the driving trajectory and determined as the second endpoint moment. The moment when the vehicle moves forward a preset time from the second endpoint moment is taken as the second start moment. The time from the second start moment to the second endpoint moment is taken as a lane change scenario.

[0014] Extract the state data of the target vehicle and its surrounding vehicles in the lane change scenario and the lane keeping scenario from the CitySim dataset, and determine the sample dataset based on the state data;

[0015] The initial lane change intention prediction model is trained based on the sample dataset to obtain a fully trained lane change intention prediction model.

[0016] In some possible implementations, state data of the target vehicle and its surrounding vehicles in the lane-changing scenario and the lane-keeping scenario are extracted from the CitySim dataset, and a sample dataset is determined based on the state data, including:

[0017] Extract the initial state data of the target vehicle and its surrounding vehicles in the lane-changing scenario and the lane-keeping scenario from the CitySim dataset;

[0018] The initial state data is normalized to obtain the final state data, and the sample dataset is determined based on the final state data.

[0019] In some possible implementations, state data of the target vehicle and its surrounding vehicles in the lane-changing scenario and the lane-keeping scenario are extracted from the CitySim dataset, including:

[0020] Based on the sliding window method, the state data of the target vehicle and its surrounding vehicles in the lane change scenario and the lane keeping scenario are extracted from the CitySim dataset.

[0021] In some possible implementations, the lane-changing scenarios include left lane-changing scenarios and right lane-changing scenarios.

[0022] In some possible implementations, the surrounding vehicles include the vehicles in front of and behind the target vehicle in the lane where the target vehicle is located, the left front vehicle, the right front vehicle, and the right rear vehicle in the lane to the left of the target vehicle, and the right front vehicle, the right vehicle, and the right rear vehicle in the lane to the right of the target vehicle.

[0023] In some possible implementations, the state data includes historical state data of the target vehicle and historical interaction data with surrounding vehicles.

[0024] In some possible implementations, the method further includes: when the lane-changing intention is lane keeping, controlling the target vehicle to execute a following strategy, the following strategy employing an adaptive cruise control following model.

[0025] In some possible implementations, the cooperative lane-changing strategy includes a two-stage cooperative strategy, which includes: based on a trajectory planning strategy, controlling the target vehicle and the cooperating vehicles in the target lane to adjust their speeds until the distance between the target vehicle and the surrounding vehicles meets the lane-changing requirements, and controlling the target vehicle to change lanes.

[0026] The vehicle trajectory in the trajectory planning strategy is described by the following formula:

[0027]

[0028]

[0029]

[0030]

[0031]

[0032] In the formula, This indicates the longitudinal displacement of the vehicle. Indicates the time for adjusting vehicle spacing. Denotes undetermined coefficients. The first three items represent the initial motion state of the vehicle. The last two terms represent the target motion state of the vehicle. ;

[0033] The objective function for the trajectory planning strategy is:

[0034]

[0035] In the formula, For the target vehicle, For collaborative vehicles, The target speed of the target vehicle. For the target speed of the cooperating vehicles, , , They are respectively , and The weighting coefficients, For the vehicle's desired speed, The vehicle's acceleration threshold. For the target vehicle acceleration at any moment For collaborative vehicles Acceleration at any moment;

[0036] The constraints of the objective function are:

[0037]

[0038] In the formula, For collaborative vehicles, For the target vehicle, for The longitudinal displacement of the vehicle. for The longitudinal displacement of the vehicle in front in the same lane is predicted based on a uniform velocity model. for Vehicle length, for The length of the vehicle in front of the vehicle in this lane, for vehicle acceleration at any moment For acceleration threshold, For safety margin, express time The longitudinal displacement of the vehicle. for The distance between the target vehicle and surrounding vehicles at any given time. The safe distance between the target vehicle and surrounding vehicles is determined by the initial and final states of the vehicles during the lane-changing phase.

[0039] In some possible implementations, the cooperative lane-changing strategy includes a parallel cooperative strategy, which includes controlling the target vehicle and the cooperative vehicle to change lanes in parallel to one side when the lane-changing space of the cooperative vehicle on the side away from the target vehicle in the target lane of the target vehicle is greater than a threshold.

[0040] The beneficial effects of this application are as follows: The intelligent connected vehicle control method for human-machine mixed driving traffic flow in the expressway weaving area provided by this application constructs a lane-changing intention prediction model based on convolutional neural network (CNN), long short-term memory network (LSTM) and attention mechanism for complex traffic scenarios in the weaving area. By fusing the interaction data of the target vehicle and its surrounding vehicles through the lane-changing intention prediction model, the lane-changing intention of intelligent connected vehicles in the expressway weaving area is effectively and accurately predicted, and the vehicle is controlled to execute the corresponding strategy, thereby improving the traffic efficiency and safety of human-machine mixed driving in the expressway weaving area. Attached Figure Description

[0041] Figure 1 A schematic flowchart of an embodiment of the intelligent connected vehicle control method for human-machine mixed driving traffic flow in the expressway weaving zone provided in this application;

[0042] Figure 2 A schematic diagram of the overall structure of the lane change intention prediction model provided in this application;

[0043] Figure 3 A schematic flowchart of an embodiment of the training steps for the lane change intention prediction model provided in this application;

[0044] Figure 4 This application provides a schematic diagram of a lane-changing scenario.

[0045] Figure 5 The lane keeping scenario diagram provided in this application;

[0046] Figure 6 This is a schematic diagram illustrating the working principle of the sliding time window method provided in this application;

[0047] Figure 7 A schematic diagram illustrating a driving scenario where the vehicle changes lanes to the left, as provided in this application;

[0048] Figure 8 For this application Figure 3 A schematic flowchart of an embodiment of step S304. Detailed Implementation

[0049] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0050] It should be understood that the illustrative drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may be implemented out of order, and steps without logical contextual relationships may be reversed or performed simultaneously. Furthermore, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor systems and / or microcontroller systems.

[0051] The terms "first," "second," etc., used in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature specified with "first" or "second" may explicitly or implicitly include at least one of those features. "And / or" describes the relationship between related objects, indicating that three relationships may exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone.

[0052] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0053] This application provides a method for intelligent connected vehicle control in traffic flow involving human and machine vehicles in weaving areas of expressways, which will be described in detail below.

[0054] Figure 1 A schematic flowchart of an embodiment of the intelligent connected vehicle control method for human-machine mixed driving traffic flow in the expressway weaving area provided in this application is shown below. Figure 1 As shown, the intelligent connected vehicle control method for mixed-driving traffic flow in expressway weaving zones includes:

[0055] S101. Obtain historical status data of the target vehicle and its surrounding vehicles;

[0056] S102. Input the historical state data into the fully trained lane change intention prediction model to obtain the lane change intention of the target vehicle. The fully trained lane change intention prediction model includes an input layer, a CNN layer, an LSTM layer, an Attention layer and an output layer connected in sequence.

[0057] It should be noted that the overall structure of the lane change intention prediction model is as follows: Figure 2 As shown, when processing sample data, the prediction model first uses a one-dimensional convolutional neural network (1D CNN) layer for feature extraction. In 1D CNN, the convolution operation is performed unidirectionally along the time axis. In this way, the model can capture local temporal features in the sequence data. Subsequently, the pooling layer performs dimensionality reduction on the output of the convolutional layer to reduce the number of parameters and extract more critical features. The advantage of the CNN layer is that it can identify local patterns in the data and preserve the temporal characteristics of the features. Compared with LSTM, CNN is more efficient in extracting local features. In addition, by using convolutional kernels of different sizes, CNN can extract features at different scales from the data, which provides the model with richer information and helps to better understand and predict complex patterns in the sequence data.

[0058] Then, the extracted features are input into the LSTM layer for sequence modeling. The LSTM network can learn important information from features at different locations and can process long sequence data. In this application, the role of the LSTM layer is to extract the temporal information of vehicle lane-changing features and learn the correlation features between sequences. The last hidden state of the LSTM layer is then passed to the attention layer for further analysis.

[0059] Next, the Attention layer enhances the model's ability to identify the importance of features by assigning different weights to different features. This mechanism allows the model to focus its attention on features that are more critical to the classification task, thereby improving the model's prediction accuracy and robustness.

[0060] Finally, the features processed by the attention layer pass through a fully connected layer and are then processed by the Softmax activation function to output the final classification probability. The Softmax layer converts the model's output into a probability distribution, which can calculate the probability of the vehicle turning left, turning right, or staying in the current lane.

[0061] S103. When the lane-changing intention is to change lanes, control the target vehicle and the cooperating vehicles in the target lane to execute a coordinated lane-changing strategy.

[0062] S104. When the lane change intention is lane keeping, control the target vehicle to execute the following strategy.

[0063] It should be noted that the following strategy of intelligent connected vehicles adopts the adaptive cruise control following model.

[0064] Compared with existing technologies, this application constructs a lane-change intention prediction model based on convolutional neural network (CNN), long short-term memory network (LSTM) and attention mechanism for complex traffic scenarios in weaving areas. By fusing the interaction data of the target vehicle and its surrounding vehicles, the lane-change intention prediction model effectively and accurately predicts the lane-change intention of intelligent connected vehicles in expressway weaving areas and controls the vehicle to execute the corresponding strategy, thereby improving the efficiency and safety of human-machine mixed driving traffic in expressway weaving areas.

[0065] To obtain a better lane change intention prediction model, in some embodiments, such as Figure 3 As shown, the lane change intention prediction model in step S102 is obtained through the following steps:

[0066] S301. Obtain the CitySim dataset and extract the driving trajectory of each vehicle from the CitySim dataset;

[0067] It should be noted that lane change intention prediction is a classification problem, which aims to predict lane change intentions within an observation window. The target vehicle and surrounding vehicles' status data are used to estimate the vehicle's position within the prediction window. In this embodiment, the probability of a lane-changing operation occurring is... , The value of determines the maximum prediction range of the prediction model. Setting it to 5 seconds allows for the evaluation of predictive models with a longer prediction range while ensuring that a sufficient number of data samples are extracted from the selected dataset.

[0068] It should also be noted that the study specifically uses Expressway A, a weaving area of ​​a domestic city, from the CitySim dataset as the research object. With a large amount of data, it can more realistically reflect the complex traffic situation of the current expressway weaving area in China.

[0069] S302. Extract the lane keeping trajectory from the driving trajectory where the lane has not changed, take the last moment of the lane keeping trajectory as the first end moment, take the moment before the first end moment as the first start moment, and take the first start moment to the first end moment as a lane keeping scenario.

[0070] S303. Extract the moment when the vehicle crosses the lane line from the driving trajectory and determine it as the second endpoint moment. Take the moment that is a preset time before the second endpoint moment as the second start moment. Take the second start moment to the second endpoint moment as a lane change scenario.

[0071] It should be noted that the lane-changing and lane-keeping scenarios are a set of time series from a single vehicle, with time values ​​ranging from 0 to... , and then from arrive The time series includes 2 seconds of historical trajectory data and 5 seconds of future trajectory data; lane-changing scenarios such as... Figure 4 As shown, in lane-changing scenarios, all samples are labeled as RLC (Right Lane Change) or LLC (Left Lane Change) based on the direction of the operation at the end of the lane-changing scenario. Whether it's a left or right lane change depends on the vehicle's direction of travel and the change in the lane it belongs to; lane-keeping scenarios are as follows... Figure 5 As shown, the lane keeping scenario is defined as the lane change time (TTLC) of a single vehicle. A set of data samples.

[0072] S304. Based on the sliding window method, extract the state data of the target vehicle and its surrounding vehicles in lane changing and lane keeping scenarios from the CitySim dataset, and determine the sample dataset based on the state data.

[0073] It should be noted that the sliding window method is a commonly used method for processing time-series data, and its working principle is as follows: Figure 6 As shown, Figure 6 middle This refers to the length of the sliding time window. This represents the sampling time interval; the sliding time window slides along the time axis each time. Length, and extract a fixed time length as In this embodiment, a 2-second sliding window is used to sample the vehicle trajectory, with a sampling time interval of 0.2 seconds, resulting in a sample set of multiple lane-change trajectory sequences.

[0074]

[0075]

[0076] In the formula, Represents a sample set of trajectory sequences. Indicates the first A trajectory sequence sample, This represents the total number of sequence samples extracted. This represents the features corresponding to the sequence. This indicates the label corresponding to the predicted sequence sample, including left lane change, right lane change, and lane keeping. 0 indicates lane keeping, 1 indicates right lane change, and 2 indicates left lane change.

[0077] S305. Train the initial lane change intention prediction model based on the sample dataset to obtain a fully trained lane change intention prediction model.

[0078] Considering that lane changing is a strategy adopted by drivers to obtain a better driving experience, this behavior involves changing from the current lane to another lane, usually accompanied by a certain lane-changing speed to ensure the continuity and safety of driving. The reasons why drivers intend to change lanes are varied, mainly including their own vehicle's driving status, the driving behavior of surrounding vehicles, and the traffic environment.

[0079] For example Figure 7 Taking the scenario of a vehicle changing lanes to the left as an example, the target vehicle TV is traveling at a higher speed, while the preceding vehicle PV is traveling at a slower speed. To seek a faster speed and a better driving experience, the target vehicle TV will intend to change lanes. In the left lane, the left preceding vehicle LPV and the left rear vehicle LFV are far apart, and the gap between them meets the safe lane-changing distance and the lane-changing conditions are met. The vehicle may have a lateral driving intention to change lanes to the left. At the same time, the vehicle TV is close to the preceding vehicle PV. To avoid a collision with the preceding vehicle, the vehicle may have a longitudinal driving intention to decelerate and change lanes. If the left rear vehicle LFV has a large acceleration, the lane-changing gap provided by the left lane will gradually decrease, and the safe lane-changing conditions will not be met. The target vehicle TV will then maintain straight driving after comprehensive consideration.

[0080] Therefore, when deciding whether to change lanes, drivers consider multiple factors, among which positional distance, speed, and surrounding traffic information are important indicators. This information helps drivers assess the relative position between the current lane and the target lane, as well as the dynamic changes that may be encountered during the lane change. Thus, in some embodiments, the state data includes the target vehicle's historical state information and its historical interactions with surrounding vehicles. The target vehicle's historical state data includes its speed, position, and lane-related data. The target vehicle's historical interaction information with surrounding vehicles includes the relative speed and relative distance between the target vehicle and surrounding vehicles. Surrounding vehicles include the vehicles in front of and behind the target vehicle in its lane, the left-front, right, and left-rear vehicles in the target vehicle's left lane, and the right-front, right, and right-rear vehicles in the target vehicle's right lane.

[0081] Considering that state data includes features such as velocity, acceleration, and distance, which have different dimensions and units, and that these dimensional differences can affect data analysis results, some embodiments perform data normalization to eliminate this effect. After normalization, all features are adjusted to the same order of magnitude. This not only helps the model more accurately capture the relative changes between features but also helps improve the convergence speed of model training. Specifically, for example... Figure 8 As shown, step S304 includes:

[0082] S801. Extract the initial state data of the target vehicle and its surrounding vehicles in lane changing and lane keeping scenarios from the CitySim dataset.

[0083] S802. Normalize the initial state data to obtain the final state data, and determine the sample dataset based on the final state data.

[0084] It should be noted that in this embodiment, a linear normalization method is used to limit the range of feature values ​​to [0,1]. To ensure the stability and generalization ability of the model training, the normalization parameters are calculated using only the statistical data of the training set. Subsequently, these parameters are applied to the data of the validation set and the test set to ensure the consistency of the entire data preprocessing process. The specific calculation formula is as follows:

[0085]

[0086]

[0087]

[0088] In the formula, This indicates the first input data in the training set. The minimum value of each feature. This indicates the first input data in the training set. The maximum value of each feature, This indicates the first input data of the validation set. One characteristic, This indicates the first input data of the test set. One characteristic.

[0089] In this embodiment, the structure and input / output of each layer of the prediction model are described:

[0090] Input Layer: In the input layer, the historical state sequences of the target vehicle and the surrounding vehicles are fused into the input feature sequence. The input is a three-dimensional vector with the following three dimensions: batch size (batch_size), input sequence length (in_seq_len), and number of features (in_channels). In this paper, batch_size = 64; frame rate (FPS) = 5, so in_seq_len = FPS * 5 * 2 = 10; in_channels = 18. Therefore, the shape of the input vector of the input layer is (64, 10, 18).

[0091] CNN Layer: In this layer, the input vector from the input layer first enters a one-dimensional convolutional layer. The number of output channels of the convolutional layer is out_channels=36, and the kernel size is kernel_size=3. Therefore, the length of the output sequence of the convolution operation is out_seq_len=in_seq_len-kernel_size+1=10-3+1=8. Thus, the shape of the vector after the one-dimensional convolutional layer is (64, 8, 36). The vector then enters a max pooling layer with a kernel size of kernel_size=3. Therefore, the length of the output sequence of the pooling operation is out_seq_len=8-3+1=6. Thus, the final shape of the output vector of the CNN layer is (64, 6, 36).

[0092] LSTM layer: This layer takes the output vector from the CNN layer as input, with input feature dimension input_size=36 and hidden state dimension hidden_size=512. Information is passed and updated between different time steps through gating mechanisms such as input gate, forget gate, and output gate to generate the hidden state output at each time step. Finally, the output vector of the LSTM layer has the shape of (64, 6, 512).

[0093] Attention Layer: This layer takes the output vector from the LSTM layer as input, where batch_size = 64, seq_len = 6, and hidden_size = 512. First, the input is linearly transformed by a fully connected layer to generate a query vector, a key vector, and a value vector, all with shapes (batch_size, seq_len, hidden_size). Then, the query, key, and value are divided into multiple heads, with the number of heads (num_heads) set to 32. Therefore, the dimension of each head is head_dim = hidden_size / num_heads = 512 / 32 = 16. Next, the dot product of the query vector and the key vector is calculated to obtain the attention scores. Then, the scores are normalized using the softmax function to obtain the attention weights, which have a shape of (batch_size, num_heads, seq_len, seq_len), i.e., (64, 32, 6, 6). The attention weights are then applied to the value vector to obtain a weighted output with the shape (batch_size, num_heads, seq_len, head_dim), which is (64, 32, 6, 16). Finally, the outputs from the multiple heads are concatenated to restore the original shape, resulting in a final output shape of (batch_size, seq_len, hidden_size), which is (64, 6, 512). The Attention layer, when processing time series data, can weight the relationship between each position in the input sequence and other positions, enabling the model to better capture global dependencies.

[0094] Output layer: In sequence models, we are typically interested in the final state of the sequence. Therefore, we extract the output of the last time step, resulting in an output vector with shape (batch_size, hidden_size), which is (64, 512). This vector is then input into a fully connected layer and normalized using softmax, ultimately yielding the output of the entire model with shape (64, 3), where 64 represents the batch size and 3 represents the number of classification categories (left lane change, right lane change, lane keeping).

[0095] In order to determine better lane-changing strategies and improve the efficiency and safety of traffic flow in the expressway weaving area where people and vehicles are mixed, in some embodiments, the cooperative lane-changing strategy includes a two-stage cooperative strategy, which includes: controlling the target vehicle and the cooperating vehicle in the target lane to adjust their speeds until the distance between the target vehicle and the surrounding vehicles meets the lane-changing requirements, and controlling the target vehicle to change lanes.

[0096] It should be noted that the cooperating vehicle can be a following vehicle or a preceding vehicle in the target lane, and it must be an intelligent connected vehicle or a connected vehicle of a natural person willing to cooperate.

[0097] In the trajectory planning strategy, the vehicle trajectory adopts a time-related approach. A quartic polynomial description:

[0098]

[0099] In the formula, This indicates the longitudinal displacement of the vehicle. Indicates the time for adjusting vehicle spacing. Let represent the coefficients to be determined, and let The initial and final state boundary constraints are represented in matrix form. ,in:

[0100]

[0101]

[0102] In the formula, The first three items represent the initial motion state of the vehicle. The last two terms represent the target motion state of the vehicle. To improve lane change smoothness, the target acceleration is increased. ;

[0103] Select , Let these be the decision variables. Construct an optimization problem, and the optimal coefficients satisfy: ;

[0104] The objective function to be optimized is:

[0105]

[0106] In the formula, For the target vehicle, For collaborative vehicles, The target speed of the target vehicle. For the target speed of the cooperating vehicles, , , They are respectively , and The weighting coefficients, For the vehicle's desired speed, The vehicle's acceleration threshold. For the target vehicle acceleration at any moment For collaborative vehicles Acceleration at time step; in optimizing the standard function, When the project enters the lane-changing phase , The speed should be close to the desired speed to reduce the impact of cooperative behavior on upstream traffic. The item aims to minimize the spacing adjustment time. The term is a penalty term, which aims to prevent the optimal solution from falling on the feasible region boundary, thereby reducing the impact of human-driven vehicle disturbances on the feasibility of rolling time-domain programming.

[0107] The constraints of the objective function are:

[0108]

[0109] In the formula, For collaborative vehicles, For the target vehicle, for The longitudinal displacement of the vehicle. for The longitudinal displacement of the vehicle in front in the same lane is predicted based on a uniform velocity model. for Vehicle length, for The length of the vehicle in front of the vehicle in this lane, for vehicle acceleration at any moment For acceleration threshold, For safety margin, express time The longitudinal displacement of the vehicle, the constraint meaning of the first three lines of the formula is: during the spacing adjustment process, C1 and C2 will not collide with the vehicle in front in this lane, and the comfort requirements will be met; for The distance between the target vehicle and surrounding vehicles at any given time. The safe distance between the target vehicle and surrounding vehicles is determined by the initial and final states of the vehicle during the lane-changing phase. The constraint meaning of the last line of the formula is: at the end of the spacing adjustment phase, the distance between the target vehicle and surrounding vehicles must meet the minimum safe distance requirement for lane changing, so as to ensure that the trajectory planning during the lane-changing phase has a solution.

[0110] Furthermore, the lane-changing strategy also includes a parallel cooperation strategy, which includes: when the lane-changing space of the cooperating vehicle on the side of the target vehicle's target lane is greater than a threshold, controlling the target vehicle and the cooperating vehicle to change lanes to one side in parallel.

[0111] It should be noted that lane change space greater than the threshold indicates that lane change space is sufficient.

[0112] It should also be noted that the parallel coordination strategy has a higher priority than the two-stage coordination strategy. That is, it first determines whether the lane-changing space of the cooperating vehicle on the side of the target vehicle away from the target vehicle is greater than a threshold. If the condition is met, the parallel coordination strategy is executed; otherwise, the two-stage coordination strategy is executed.

[0113] The above provides a detailed description of the intelligent connected vehicle control method for human-machine mixed driving traffic flow in expressway weaving areas provided by this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

[0114] The above description is merely a preferred embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for intelligent connected vehicle control in traffic flow involving human-machine hybrid driving in expressway weaving zones, characterized in that, include: Acquire historical status data of the target vehicle and its surrounding vehicles; The historical state data is input into a fully trained lane change intention prediction model to obtain the lane change intention of the target vehicle. The fully trained lane change intention prediction model includes an input layer, a CNN layer, an LSTM layer, an Attention layer, and an output layer connected in sequence. When the lane-changing intention is to change lanes, control the target vehicle and cooperating vehicles in the target lane to execute a coordinated lane-changing strategy; The cooperative lane-changing strategy includes a two-stage cooperative strategy, which includes: based on a trajectory planning strategy, controlling the target vehicle and the cooperating vehicles in the target lane to adjust their speeds until the distance between the target vehicle and the surrounding vehicles meets the lane-changing requirements, and then controlling the target vehicle to change lanes. The vehicle trajectory in the trajectory planning strategy is described by the following formula: X In the formula, This indicates the longitudinal displacement of the vehicle. Indicates the time for adjusting vehicle spacing. Denotes undetermined coefficients. The first three items represent the initial motion state of the vehicle. The last two terms represent the target motion state of the vehicle. ; The objective function for the trajectory planning strategy is: In the formula, For the target vehicle, For collaborative vehicles, The target speed of the target vehicle. For the target speed of the cooperating vehicles, , , They are respectively , and The weighting coefficients, For the vehicle's desired speed, The vehicle's acceleration threshold. For the target vehicle acceleration at any moment For collaborative vehicles acceleration at any moment The item is to enable the entry into the lane changing phase. , The speed should be close to the desired speed to reduce the impact of cooperative behavior on upstream traffic. To minimize the spacing adjustment time, The term is a penalty term, which aims to prevent the optimal solution from falling on the feasible region boundary, thereby reducing the impact of human-driven vehicle disturbances on the feasibility of rolling time-domain programming. The constraints of the objective function are: In the formula, The longitudinal displacement of the vehicle. for The longitudinal displacement of the vehicle in front in the same lane is predicted based on a uniform velocity model. for Vehicle length, for The length of the vehicle in front of the vehicle in this lane, for vehicle acceleration at any moment For acceleration threshold, For safety margin, The distance between the target vehicle and surrounding vehicles at any given time. The safe distance between the target vehicle and surrounding vehicles is determined by the initial and final states of the vehicles during the lane-changing phase.

2. The intelligent connected vehicle control method for mixed human-machine driving traffic flow in expressway weaving areas according to claim 1, characterized in that, The fully trained lane change intention prediction model is obtained through the following steps: Obtain the CitySim dataset and extract the driving trajectory of each vehicle from the CitySim dataset; Extract lane-keeping trajectories from the driving trajectory where the lane does not change, take the last moment of the lane-keeping trajectory as the first end moment, take the moment before the first end moment as the first start moment, and take the time from the first start moment to the first end moment as a lane-keeping scenario. The moment when the vehicle crosses the lane line is extracted from the driving trajectory and determined as the second endpoint moment. The moment when the vehicle moves forward a preset time from the second endpoint moment is taken as the second start moment. The time from the second start moment to the second endpoint moment is taken as a lane change scenario. Extract the state data of vehicles and surrounding vehicles in the lane-changing scenario and the lane-keeping scenario from the CitySim dataset, and determine the sample dataset based on the state data; The initial lane change intention prediction model is trained based on the sample dataset to obtain a fully trained lane change intention prediction model.

3. The intelligent connected vehicle control method for mixed human-machine driving traffic flow in expressway weaving areas according to claim 2, characterized in that, State data of the target vehicle and its surrounding vehicles in the lane-changing and lane-keeping scenarios are extracted from the CitySim dataset. Based on the state data, a sample dataset is determined, including: Extract the initial state data of the target vehicle and its surrounding vehicles in the lane-changing scenario and the lane-keeping scenario from the CitySim dataset; The initial state data is normalized to obtain the final state data, and the sample dataset is determined based on the final state data.

4. The intelligent connected vehicle control method for mixed human-machine driving traffic flow in expressway weaving areas according to claim 2, characterized in that, Extract the state data of the target vehicle and its surrounding vehicles in the lane-changing scenario and the lane-keeping scenario from the CitySim dataset, including: Based on the sliding window method, the state data of the target vehicle and its surrounding vehicles in the lane change scenario and the lane keeping scenario are extracted from the CitySim dataset.

5. The intelligent connected vehicle control method for mixed human-machine driving traffic flow in expressway weaving areas according to claim 2, characterized in that, The lane-changing scenarios include left lane-changing scenarios and right lane-changing scenarios.

6. The intelligent connected vehicle control method for mixed human-machine driving traffic flow in expressway weaving areas according to claim 1, characterized in that, The surrounding vehicles include the vehicles in front of and behind the target vehicle in the lane where the target vehicle is located, the left front vehicle, the left vehicle and the left rear vehicle in the left lane of the target vehicle, and the right front vehicle, the right vehicle and the right rear vehicle in the right lane of the target vehicle.

7. The intelligent connected vehicle control method for mixed human-machine driving traffic flow in expressway weaving areas according to claim 1, characterized in that, The status data includes the target vehicle's historical status data and its historical interaction data with surrounding vehicles.

8. The intelligent connected vehicle control method for mixed human-machine driving traffic flow in expressway weaving areas according to claim 1, characterized in that, The method further includes: when the lane change intention is lane keeping, controlling the target vehicle to execute a following strategy, wherein the following strategy adopts an adaptive cruise control following model.

9. The intelligent connected vehicle control method for mixed human-machine driving traffic flow in expressway weaving areas according to claim 1, characterized in that, The cooperative lane-changing strategy includes a parallel cooperative strategy, which includes: when the lane-changing space of the cooperating vehicle on the side of the target vehicle's target lane is greater than a threshold, controlling the target vehicle and the cooperating vehicle to change lanes to one side in parallel.

Images