A lane-changing guidance method for vehicles in highway tunnels based on travel time prediction model

Abstract

This invention belongs to the field of intelligent transportation and discloses a method for guiding vehicle lane changing in highway tunnels based on a travel time prediction model, including: S1. acquiring trajectory data of intelligent connected vehicles in the tunnel and preprocessing the trajectory data; S2. performing correlation analysis on travel time characteristics, using T... pre The invention extracts travel time features using seconds as the detection period, normalizes the data, and constructs a travel time prediction dataset. Step S3 involves building a PSO-CNN-LSTM-AM travel time prediction model, inputting the travel time prediction dataset into the model for training, and then conducting experimental comparisons and verifications. Step S4 utilizes the model trained in step S3 for periodic predictions, and based on the predicted travel time values, proposes a lane-changing control method for highway tunnel entrance areas. This invention can better extract important features, achieve more accurate prediction results, improve the utilization rate of two-lane tunnels and traffic efficiency, and alleviate traffic congestion.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation, specifically relating to a method for guiding vehicle lane changes in highway tunnels based on a travel time prediction model. Background Technology

[0002] Travel time prediction is the forecast of the time required to traverse a specific road segment or route within a future timeframe. Accurate predictions reflect changing trends in road traffic and vehicle operation, which is crucial for dynamic traffic management in highway tunnels. Furthermore, with the development of intelligent connected vehicles, the integration of advanced sensors, control systems, actuators, and communication and network technologies enables real-time information exchange, effectively addressing traffic congestion and safety challenges in highway tunnels. Therefore, considering the inherent structural and environmental factors, inconsistent speeds in fast and slow lanes, and diverse vehicle types in existing highway tunnels, which often lead to uneven lane utilization and low traffic efficiency, it is necessary to design a dual-lane travel time prediction method for tunnels based on intelligent connected vehicle trajectory data, as well as a lane-changing guidance method for highway tunnels based on travel time prediction values. This will allow for the rational allocation of traffic flow and improved tunnel throughput.

[0003] A review of relevant patents and papers reveals that current travel time prediction methods primarily rely on macro-level data such as toll data. Patent CN106228268A, based on toll data, establishes a travel time prediction model combining an autoencoder and a backpropagation neural network, improving the model's adaptability and stability. Patent CN113065684A, based on highway toll data, establishes a highway travel time prediction method combining a VAE and deep learning model, improving prediction accuracy. However, these methods, primarily based on macro-level data, suffer from insufficient generalization ability and prediction accuracy, and their long prediction times make them less suitable for real-time traffic control strategies. Furthermore, existing methods using trajectory data for prediction lack sufficient extraction of key features. Patent CN106228808A, based on floating car data, establishes a method for predicting travel time on urban expressways, influencing factors such as weekdays / non-weekdays, historical travel time, and average speed, demonstrating strong robustness and fault tolerance. Patent CN115565376A, based on trajectory data, establishes a method for predicting travel time using a fusion graph2vec and two-layer LSTM approach, influencing factors such as traffic congestion, historical travel time, weather, weekdays / non-weekdays, and temperature, improving resource utilization efficiency. However, these methods lack sufficient feature mining for key aspects of travel time prediction using trajectory data. In highway tunnel scenarios, factors such as headway, the proportion of large vehicles, and the number of vehicles all have an impact. Furthermore, research on vehicle lane-changing guidance methods at highway tunnel entrances based on travel time prediction models is currently limited, both domestically and internationally. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a vehicle lane-changing guidance method for highway tunnels based on a travel time prediction model. This invention aims to solve the problems of low prediction accuracy and difficulty in rationally allocating traffic flow in existing methods, thereby improving tunnel throughput.

[0005] This invention provides a method for guiding vehicle lane changes in highway tunnels based on a travel time prediction model, comprising the following steps:

[0006] S1. Acquire trajectory data of intelligent connected vehicles in the tunnel and preprocess the trajectory data;

[0007] The trajectory data includes the speed V and position X of the intelligent connected vehicle;

[0008] S2. Conduct correlation analysis on the travel time characteristics, using T pre The travel time features are extracted using seconds as the detection period, and the data is normalized to construct a travel time prediction dataset.

[0009] S3. Build a PSO-CNN-LSTM-AM travel time prediction model, input the travel time prediction dataset into the model for training, and then conduct experiments for comparison and verification.

[0010] S4. Using the model trained in step S3, periodic predictions are made, and based on the predicted travel time, a lane-changing control method for the entrance area of ​​a highway tunnel is proposed.

[0011] Furthermore, in step S1, the step of filtering and preprocessing the trajectory data is as follows:

[0012] I. A first-order inertial filter is used to filter the n trajectory data D per second. First, the smoothing result is initialized with the first sample value, and then the filtering is performed using the following formula:

[0013] D smooth [i] = α*D[i] + (1-α)*D smooth [i-1]

[0014] In the formula, α represents the filter coefficient, α = 0.2; D smooth [i] represents the output value of the filtered data; D smooth [i-1] represents the output value of the previous filtered data; D[i] represents the value of the current sampled data;

[0015] II. Remove the highest data value D from the n data points using the median average method. high and the lowest data value D lowCalculate the average of the remaining n-2 data points as the data information D at 1-second time intervals. i D i The calculation expression is as follows:

[0016]

[0017] Furthermore, step S2 includes the following sub-steps:

[0018] S2.1 Perform autocorrelation analysis on historical travel times, and the calculation formula is as follows:

[0019]

[0020] In the formula, k is the travel time order; μ is the mean of the travel time series; T i and T i+k is the i-th term of the two subsequences split by time lag; ρ is the autocorrelation coefficient, used to quantitatively describe the strength of the correlation between travel times and has directionality. The value of ρ ranges from [-1, 1]. The closer it is to 1, the higher the positive correlation of travel times; the closer it is to -1, the lower the negative correlation of travel times; N represents the number of vehicles.

[0021] S2.2 uses Pearson correlation coefficient to analyze travel time characteristics, including the historical travel time series of the two-lane tunnel. Large vehicle proportion sequence data P, vehicle headway characteristics T L Vehicle number characteristic N and speed characteristic V;

[0022] Assume variable X = {x1, x2, x3, ..., x n}, where X represents P and T L Any one of the variables N and V;

[0023] Variable T = {T1, T2, T3, ..., T} n The formula for calculating the Pearson correlation coefficient between two variables is as follows:

[0024]

[0025] In the formula, r represents the correlation coefficient between the two variables; Cov(X,T) represents the covariance between the feature variable X and the travel time T; Var(X) and Var(T) represent the sample variances of the feature variable X and the travel time T, respectively.

[0026] S2.3 Extract travel time features with T=k as the detection period and normalize the data;

[0027] The maximum-minimum normalization method is used to normalize the travel time data and feature data in the original dataset, converting them into values ​​within the range of [0,1]. The calculation expression is as follows:

[0028]

[0029] In the formula, X Norm This represents the standardized data of travel time and feature data; X chushi This represents the original travel time and feature data; min(X) and max(X) are the minimum and maximum values ​​in the original data, respectively.

[0030] S2.4 Construct a travel time prediction dataset and divide it into training and test sets.

[0031] Furthermore, step S3 includes the following sub-steps:

[0032] S3.1 Construct a CNN layer to capture travel time and vehicle trajectory data features. Use one-dimensional convolution with a kernel size of 3, a filter number of 64, and the activation function ReLU.

[0033] S3.2 An LSTM layer is superimposed to capture the autocorrelation of travel time series and the short-term and long-term characteristics of travel time series data. A Dropout layer is introduced to prevent overfitting and is set to 0.3.

[0034] S3.3 Optimize the LSTM learning rate alpha and the number of neurons in the first hidden layer hidden_nodes using the PSO optimization function;

[0035] S3.4 introduces an AM layer, uses an additive model to obtain attention scores, stacks a Dense layer, uses the Softmax activation function to output weight values, multiplies and weights the LSTM output to obtain the trip time prediction output value;

[0036] S3.5 performs inverse normalization on the predicted travel time output and conducts experimental comparison and verification.

[0037] Furthermore, step S4 includes the following sub-steps:

[0038] S4.1 Utilize the PSO-CNN-LSTM-AM travel time prediction model established in step S3, with T pre The travel time for each lane inside the tunnel is predicted in seconds.

[0039] S4.2 Analysis of the travel time difference threshold ΔT for lane-changing strategies at the tunnel entrance area threshold And determine the difference in predicted travel time between the two lanes, ΔTT. preDoes it exceed the difference threshold ΔT? threshold ;

[0040] If the threshold is not exceeded, the process is repeated to predict the travel time at the next moment.

[0041] If the threshold is exceeded, the intelligent connected vehicle lane-changing strategy is activated. If the lane-changing strategy is activated, the relationship between the travel time difference and the optimal number of lane changes in the highway tunnel is analyzed, and the corresponding optimal number of lane changes N is calculated. change ;

[0042] S4.3 during lane change control cycle T C Within the tunnel, the optimal number of lane changes N is satisfied at the entrance area of ​​the highway tunnel. change Intelligent connected vehicles can change lanes from lanes with longer travel times to lanes with shorter travel times, and then periodically predict and implement intelligent connected vehicle lane-changing control strategies.

[0043] Furthermore, in step S4.2, the threshold value ΔT for the travel time difference is analyzed. threshold A multi-objective function, which evaluates the traffic efficiency after implementing the lane-changing strategy, is selected based on average travel time, average lost time, and average fuel consumption. The objective function is as follows:

[0044]

[0045] In the formula, β1, β2, and β3 represent the average travel time, respectively. Average Loss Time Average fuel consumption The correlation coefficients are β1 = β2 = β3 = 1.

[0046] Furthermore, in step S4.2, the optimal number of lane changes N change The calculation steps are as follows:

[0047] Using average travel time Average Loss Time Average fuel consumption The objective function is used to analyze the optimal number of lane changes. By minimizing the objective function, the difference in travel time prediction ΔTT for each trip is obtained. pre The optimal number of lane changes;

[0048]

[0049] In the formula,

[0050] Beneficial effects:

[0051] Based on the trajectory data of intelligent connected vehicles on the road, this paper fully explores the influence of multiple features such as historical travel time, the proportion of large vehicles, headway, number of vehicles in each lane, and average speed. It presents a PSO-CNN-LSTM-AM method for predicting travel time in highway tunnels, which can better extract important features and achieve more accurate prediction results. Simultaneously, based on the difference in travel time prediction between two lanes, this paper analyzes the difference threshold and the optimal number of lane changes, providing a method for guiding intelligent connected vehicles to change lanes in the entrance area of ​​highway tunnels using the difference in travel time between two lanes. This method can improve the utilization rate of the two lanes in the tunnel and traffic efficiency, alleviating traffic congestion.

[0052] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0053] Figure 1 A schematic diagram illustrating the steps involved in predicting travel time through a highway tunnel.

[0054] Figure 2 This is a schematic diagram of the model structure for a travel time prediction method based on PSO-CNN-LSTM-AM.

[0055] Figure 3 This is a flowchart of a lane-changing control method for highway tunnel entrance areas based on a two-lane travel time prediction model.

[0056] Figure 4 This is a schematic diagram of a two-lane highway tunnel scenario in this embodiment;

[0057] Figure 5 A schematic diagram illustrating the changes in traffic efficiency under different travel time difference thresholds;

[0058] Figure 6 This is a schematic diagram of the optimal lane change number prediction model based on random forest in this embodiment;

[0059] Figure 7 This is a comparison chart of the predicted and actual travel time for the two-lane tunnel in this embodiment;

[0060] Figure 8 This is a comparison chart of the predicted and actual travel times for two lanes with and without CAV lane-changing control strategies when the CAV penetration rate is 60% in this embodiment. Detailed Implementation

[0061] To make the technical solutions, advantages, and objectives of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the protection scope of this application.

[0062] like Figure 1 As shown, this invention provides a method for guiding vehicle lane changes in highway tunnels based on a travel time prediction model, comprising the following steps:

[0063] S1. Acquire trajectory data of intelligent connected vehicles in the tunnel and preprocess the trajectory data;

[0064] The trajectory data of intelligent connected vehicles in the tunnel may have large fluctuations. Therefore, it is necessary to smooth the trajectory data D (including speed V and position X), filter the n data points per second, and finally keep only one data point per second.

[0065] In step S1, the filtering preprocessing step for the trajectory data is as follows:

[0066] I. A first-order inertial filter is used to filter the n trajectory data D per second. First, the smoothing result is initialized with the first sample value, and then the filtering is performed using the following formula:

[0067] D smooth [i] = α*D[i] + (1-α)*D smooth [i-1]

[0068] In the formula, α represents the filter coefficient, α = 0.2; D smooth [i] represents the output value of the filtered data; D smooth [i-1] represents the output value of the previous filtered data; D[i] represents the current sampled data value; the current sampled value and the previous filtered output value are weighted to obtain the effective filtered data value, so that the output has a feedback effect on the input.

[0069] II. Remove the highest data value D from the n data points using the median average method. high and the lowest data value D low Calculate the average of the remaining n-2 data points as the data information D at 1-second time intervals. i D i The calculation expression is as follows:

[0070]

[0071] S2. Conduct correlation analysis on the travel time characteristics, using T pre The travel time features are extracted using seconds as the detection period, and the data is normalized to construct a travel time prediction dataset.

[0072] S2.1 In the travel time series data, there is a certain correlation between the travel time value in the current period and its lagged historical travel time. Autocorrelation analysis is performed on the historical travel time, and the calculation formula is as follows:

[0073]

[0074] In the formula, k is the travel time order; μ is the mean of the travel time series; T i and T i+k is the i-th term of the two subsequences split by time lag; ρ is the autocorrelation coefficient, used to quantitatively describe the strength of the correlation between travel times and has directionality. The value of ρ ranges from [-1, 1]. The closer it is to 1, the higher the positive correlation of travel times; the closer it is to -1, the lower the negative correlation of travel times; N represents the number of vehicles.

[0075] S2.2 uses Pearson correlation coefficient to analyze travel time characteristics, including the historical travel time series of the two-lane tunnel. Large vehicle proportion sequence data P, vehicle headway characteristics T L Vehicle number characteristic N and speed characteristic V;

[0076] Assume variable X = {x1, x2, x3, ..., x n}, where X represents P and T L Any one of the variables N and V;

[0077] Variable T = {T1, T2, T3, ..., T} n The formula for calculating the Pearson correlation coefficient between two variables is as follows:

[0078]

[0079] In the formula, r represents the correlation coefficient between the two variables; Cov(X,T) represents the covariance between the feature variable X and the travel time T; Var(X) and Var(T) represent the sample variances of the feature variable X and the travel time T, respectively.

[0080] S2.3 Extract travel time features with T=k as the detection period and normalize the data;

[0081] To effectively improve the convergence speed and accuracy of the prediction model and prevent gradient explosion, the minimax normalization method is used to normalize the travel time data and feature data in the original dataset, converting them into values ​​within the range of [0,1]. The calculation expression is as follows:

[0082]

[0083] In the formula, X Norm This represents the standardized data of travel time and feature data; X chushi This represents the original travel time and feature data; min(X) and max(X) are the minimum and maximum values ​​in the original data, respectively.

[0084] S2.4 Construct a travel time prediction dataset and divide it into training and test sets.

[0085] S3. For example Figure 2 As shown, a PSO-CNN-LSTM-AM travel time prediction model was built. The travel time prediction dataset was input into the model for training, and then experiments were conducted for comparison and verification.

[0086] S3.1 Construct a CNN layer to capture travel time and vehicle trajectory data features. Use one-dimensional convolution with a kernel size of 3, a filter number of 64, and the activation function ReLU.

[0087] First, a CNN layer is built, which takes the historical travel time series of the tunnel's two lanes as input. Large vehicle proportion sequence data P, vehicle headway characteristics T L The vehicle number feature N and speed feature V are input into the convolutional layer, and the ReLU activation function is used:

[0088]

[0089] Where Y c Let f(z) represent the output of the convolutional layer, f(z) represent the activation function, and W represent the output of the convolutional layer. c Represents the filter weights, X represents the convolution operator. Norm B represents the standardized travel time and feature data. c This represents the filter bias value. Next, the output of the convolutional layer is connected to a pooling layer, and max pooling is used to obtain a local optimum in the convolutional feature matrix, effectively mitigating overfitting. The calculation process is as follows:

[0090] Y p =Pool max (Y c )

[0091] Where Y p Represents the output of the pooling layer, Pool max (z) represents the max pooling function.

[0092] S3.2 An LSTM layer is superimposed to capture the autocorrelation of travel time series and the short-term and long-term characteristics of travel time series data. A Dropout layer is introduced to prevent overfitting and is set to 0.3.

[0093] The pooling layer data is then flattened, and an LSTM layer is added to capture the dependencies in the travel time series data, achieving accurate prediction. The activation function is Selu, with the learning rate alpha and the number of neurons in the first hidden layer (hidden_nodes) set.

[0094] S3.3 Optimize the LSTM learning rate alpha and the number of neurons in the first hidden layer hidden_nodes using the PSO optimization function;

[0095] Next, the PSO optimization algorithm is introduced to optimize the two parameters of the LSTM layer, thereby improving the prediction and generalization ability of the LSTM model. By using the Adam optimizer, setting MSE as the loss function, and using the mean squared error between the predicted travel time and the actual travel time as the fitness function, the global optimal solution and fitness are recorded to obtain the optimal learning rate alpha and the number of neurons in the first hidden layer, hidden_nodes.

[0096] S3.4 introduces an AM layer, uses an additive model to obtain attention scores, stacks a Dense layer, uses the Softmax activation function to output weight values, multiplies and weights the LSTM output to obtain the trip time prediction output value;

[0097] Finally, an AM layer is constructed, and the attention mechanism is used to learn the weights at each LSTM time point. Softmax is used as the activation function, and the weight probability calculation formula is as follows:

[0098]

[0099] Where s(x) i ,q) represents the attention scoring function, which is calculated using an additive model:

[0100] s(x i ,q)=v T tanh(Wx i +Uq)

[0101] Where W, U, and v represent network parameters. Finally, a weighted average is used to multiply and weight the predicted travel times, resulting in the predicted travel time output value TT.pre The calculation formula is as follows:

[0102]

[0103] S3.5 performs inverse normalization on the predicted travel time output and conducts experimental comparison and verification.

[0104] TTs pre =TT pre *(max(TT pre )-min(TT pre ))+min(TT pre )

[0105] Finally, the trained tunnel travel time prediction model based on PSO-CNN-LSTM-AM was experimentally verified and compared.

[0106] S4. Using the model trained in step S3, periodic predictions are made, and based on the predicted travel time, a lane-changing control method for the entrance area of ​​a highway tunnel is proposed.

[0107] S4.1 Utilize the PSO-CNN-LSTM-AM travel time prediction model established in step S3, with T pre Using seconds as the period, the travel time is predicted for each lane inside the tunnel, and the predicted travel time for both lanes is calculated as TTs. pre1 and TTs pre2 Input into subsequent lane-changing strategies;

[0108] Based on the above-obtained two-lane travel time prediction value TTs pre1 and TTs pre2 ,like Figure 3 As shown, this invention proposes a lane-changing control method for highway tunnel entrance areas based on a travel time prediction model;

[0109] S4.2 Analysis of the travel time difference threshold ΔT for lane-changing strategies at the tunnel entrance area threshold And determine the difference in predicted travel time between the two lanes, ΔTT. pre Does it exceed the difference threshold ΔT? threshold ;

[0110] like Figure 4 As shown, the scenario depicts a two-lane highway tunnel, comprising the tunnel entrance area (intelligent connected vehicle lane change control zone), the tunnel interior (no lane change control zone), and the tunnel exit area. The analysis focuses on the travel time difference threshold ΔT. threshold A multi-objective function, which evaluates the traffic efficiency after implementing the lane-changing strategy, is selected based on average travel time, average lost time, and average fuel consumption. The objective function is as follows:

[0111]

[0112] In the formula, β1, β2, and β3 represent the average travel time, respectively. Average Loss Time Average fuel consumption The correlation coefficients are calculated. Considering the inconsistent dimensions of the data, the minimum-maximum normalization method is used for normalization. Next, considering that all three variables have a significant impact on the objective function, the correlation coefficients are set as follows: β1 = β2 = β3 = 1.

[0113] like Figure 5 As shown, the traffic efficiency changes under different travel time difference thresholds are obtained by minimizing the above objective function, thus yielding the optimal travel time difference lane-changing threshold ΔT. threshold Then, determine the difference in predicted travel time between the two lanes, ΔTT. pre Does it exceed the difference threshold ΔT? threshold .

[0114] If the threshold is not exceeded, the process is repeated to predict the travel time at the next moment.

[0115] If the threshold is exceeded, the intelligent connected vehicle lane-changing strategy is activated. If the lane-changing strategy is activated, the relationship between the travel time difference and the optimal number of lane changes in the highway tunnel is analyzed, and the corresponding optimal number of lane changes N is calculated. change ;

[0116] Number of lane changes N change Difference between travel time and prediction ΔTT pre They are closely related; as the difference between the two lanes increases, the number of lane changes also increases. Similarly, using average travel time... Average Loss Time Average fuel consumption The objective function is used to analyze the optimal number of lane changes. By minimizing the objective function, the difference in travel time prediction ΔTT for each trip is obtained. pre The optimal number of lane changes;

[0117]

[0118] Similarly, the correlation coefficients mentioned above are set as follows:

[0119] Next, random forest is used to calculate the difference in travel time prediction ΔTT. pre The relationship with the number of lane changes in CAVs is quantified, such as... Figure 6 As shown, a lane change number prediction model based on random forest is obtained.

[0120] After enabling the CAV lane-changing strategy, the optimal number of lane changes N is calculated based on the aforementioned lane-changing quantity prediction model and the travel time difference. change .

[0121] S4.3 during lane change control cycle T C Within the tunnel, the optimal number of lane changes N is satisfied at the entrance area of ​​the highway tunnel. change Intelligent connected vehicles can change lanes from lanes with longer travel times to lanes with shorter travel times, and then periodically predict and implement intelligent connected vehicle lane-changing control strategies.

[0122] Example

[0123] This embodiment uses vehicle trajectory data from the Taiping Tunnel on the Chongqing Expressway as the research object. Vehicle trajectory data from August 18th to August 21st, 2023 (four days in total) were selected as example data. The data frequency was 10 frames per second, and the training and test sets were divided in a 7:3 ratio. The travel time prediction period was set to T. pre =5.

[0124] Two-lane journey time prediction results based on PSO-CNN-LSTM-AM

[0125] like Figure 7 As shown, the travel time prediction model constructed above is used to predict the travel time of the two lanes in the tunnel. The predicted values ​​for both lanes are close to the actual values, and the prediction error is small. Figure 7 (a) is the travel time for lane 1 at a T = 5s interval. Figure 7 (b) is the travel time for the two lanes at a time interval of T = 5 seconds. To verify the effectiveness of the model, Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE) were selected as evaluation metrics.

[0126]

[0127] To verify the effectiveness of the model, it was compared with various neural network models. Table 1 below shows the prediction results of different models.

[0128] Table 1

[0129]

[0130] The results show that, compared with traditional methods for travel time prediction, the PSO-CNN-LSTM-AM highway tunnel travel time prediction model proposed in this invention, which considers the influence of multiple factors, can fully explore the spatiotemporal characteristics of the travel time process and has high prediction accuracy. Moreover, it can achieve ideal prediction results when applied to different lanes, providing a foundation for the subsequent establishment of CAV lane-changing guidance methods in tunnel entrance areas.

[0131] Next, to verify the impact of different data volumes on the prediction of travel time for two lanes, the above trajectory data was sampled. Table 2 below shows the prediction error of travel time for two lanes with sampled data volumes of 20%, 40%, 60%, and 80%.

[0132] Table 2

[0133]

[0134] The results show that as the amount of trajectory data sampled increases, the prediction errors for travel time in both lane 1 and lane 2 decrease. However, when the amount of trajectory data sampled is 20%, the MAE (Modal Analysis Error) is greater than 10, indicating poor prediction performance. This is because the collected data is limited and differs significantly from the overall traffic conditions. When the amount of trajectory data sampled reaches 40%, a more accurate prediction effect is achieved. In summary, when the amount of trajectory data sampled is greater than 20%, the model prediction results of this invention are generally consistent with the changing trends of actual travel time data, demonstrating a relatively ideal prediction effect.

[0135] Validation of the Effectiveness of a Vehicle Lane Changing Guidance Method Based on Two-Lane Travel Time Prediction at Tunnel Entrance Area

[0136] This embodiment establishes a co-simulation platform based on Sumo and Python to verify the travel time prediction method and lane change control method provided by this invention. A highway tunnel in Chongqing is selected as the test background section. The tunnel has a total length of 4.237km, with two lanes in one direction. The section before the tunnel entrance is 2000m long, the tunnel itself is 2037m long, and the section after the tunnel exit is 200m long. Taking a CAV penetration rate of 60% as an example... Figure 8 As shown, the predicted and actual travel times for two lanes are compared with and without the CAV lane change control strategy. Figure 8 (a) Comparison of actual and predicted travel time for a two-lane road without lane-changing control at a CAV penetration rate of 60%. Figure 8 (b) Comparison of actual and predicted travel times for two lanes under lane change control at 60% CAV penetration.

[0137] Finally, the traffic efficiency with and without lane-change control strategies is evaluated using total travel time, total lost time, total fuel consumption, and total traffic volume. Table 3 below is a comparison table of traffic efficiency with and without lane-change control strategies under different CAV penetration rates.

[0138] Table 3

[0139]

[0140] The results show that, compared with lane-changing control without lane changing, the lane-changing control designed in this invention can improve tunnel throughput and reduce travel time and lost time under any CAV penetration rate. Furthermore, with increasing CAV penetration, both lane-changing control without and with lane changing can improve tunnel throughput. This is because the ability to adaptively adjust congestion is enhanced when the number of CAVs increases. Additionally, higher prediction accuracy provides more precise control. In summary, the highway tunnel lane-changing control method based on a travel time prediction model proposed in this invention can accurately predict travel time and effectively improve tunnel traffic efficiency, alleviating traffic congestion.

[0141] It is hereby declared that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for guiding vehicle lane changing in highway tunnels based on a travel time prediction model, characterized in that, Includes the following steps: S1. Acquire trajectory data of intelligent connected vehicles in the tunnel and preprocess the trajectory data; The trajectory data includes the speed V and position of the intelligent connected vehicle; S2. Conduct correlation analysis on the travel time characteristics, using T pre The travel time features are extracted using seconds as the detection period, and the data is normalized to construct a travel time prediction dataset. S3. Build a PSO-CNN-LSTM-AM travel time prediction model, input the travel time prediction dataset into the model for training, and then conduct experiments for comparison and verification. S3.1 Construct a CNN layer to capture travel time and vehicle trajectory data features. Use one-dimensional convolution with a kernel size of 3, a filter number of 64, and the activation function ReLU. S3.2 An LSTM layer is superimposed to capture the autocorrelation of the travel time series and the short-term and long-term characteristics of the travel time series data. A Dropout layer is introduced to prevent overfitting and is set to 0.

3. S3.3 Optimize the LSTM learning rate alpha and the number of neurons in the first hidden layer hidden_nodes using the PSO optimization function; S3.4 introduces an AM layer, uses an additive model to obtain attention scores, stacks a Dense layer, uses the Softmax activation function to output weight values, multiplies and weights the LSTM output to obtain the trip time prediction output value; S3.5 performs inverse normalization on the predicted travel time output and conducts experimental comparison and verification. S4. Using the model trained in step S3, periodic predictions are made, and based on the predicted travel time, a lane-changing control method for the entrance area of ​​a highway tunnel is proposed. S4.1 Utilize the PSO-CNN-LSTM-AM travel time prediction model established in step S3, with T pre The travel time for each lane inside the tunnel is predicted in seconds. S4.2 Analysis of the travel time difference threshold ΔT for lane-changing strategies at the tunnel entrance area threshold And determine the difference in predicted travel time between the two lanes, ΔTT. pre Does it exceed the difference threshold ΔT? threshold ; If the threshold is not exceeded, the process is repeated to predict the travel time at the next moment. If the threshold is exceeded, the intelligent connected vehicle lane-changing strategy will be activated. If the lane-changing strategy is enabled, analyze the relationship between the difference in travel time prediction and the optimal number of lane changes in the highway tunnel, and calculate the corresponding optimal number of lane changes N. change ; S4.3 during lane change control cycle T C Within the tunnel, the optimal number of lane changes N is satisfied at the entrance area of ​​the highway tunnel. change Intelligent connected vehicles can change lanes from lanes with longer travel times to lanes with shorter travel times, and then periodically predict and implement intelligent connected vehicle lane-changing control strategies.

2. The method for guiding vehicle lane changing in highway tunnels based on a travel time prediction model according to claim 1, characterized in that: In step S1, the step of filtering and preprocessing the trajectory data is as follows: I. A first-order inertial filter is used to filter the n trajectory data D per second. First, the smoothing result is initialized with the first sample value, and then the filtering is performed using the following formula: D smooth [i]=α*D[i]+(1-α)*D smooth [i-1] In the formula, α represents the filter coefficient, α = 0.2; D smooth [i] represents the output value of the filtered data; D smooth [i-1] represents the output value of the previous filtered data; D[i] represents the value of the sampled data in this instance; II. Remove the highest data value D from the n data points using the median average method. high and the lowest data value D low Calculate the average of the remaining n-2 data points as the data information at 1-second intervals, D. i The calculation expression is as follows:

3. The method for guiding vehicle lane changing in highway tunnels based on a travel time prediction model according to claim 1, characterized in that: Step S2 includes the following sub-steps: S2.1 Perform autocorrelation analysis on historical travel times, and the calculation formula is as follows: In the formula, k is the travel time order; μ is the mean of the travel time series; T i and T i+k Let i be the i-th and i+k-th terms of the two subsequences split by time delay; ρ is the autocorrelation coefficient, which is used to quantitatively describe the strength of the correlation between travel times and has a directionality. The value of ρ ranges from [-1, 1]. The closer it is to 1, the stronger the positive correlation between travel times. The closer the value is to -1, the lower the negative correlation between travel time and travel duration. S2.2 uses Pearson correlation coefficient to analyze travel time characteristics, including the historical travel time series of the two-lane tunnel. Large vehicle proportion sequence data P, vehicle headway characteristics T L Vehicle number characteristic N′ and speed characteristic V; Assume variable X = {x1, x2, x3, ..., x n }, where X represents P and T L Any one of the variables N′ and V; Variable T = {T1, T2, T3, ..., T} n The formula for calculating the Pearson correlation coefficient between two variables is as follows: In the formula, r represents the correlation coefficient between the two variables; Cov(X,T) represents the covariance between the feature variable X and the travel time T; Var(X) and Var(T) represent the sample variances of the feature variable X and the travel time T, respectively. S2.3 with T pre The travel time features are extracted using seconds as the detection cycle, and the data is then normalized. The maximum-minimum normalization method is used to normalize the travel time data and feature data in the original dataset, converting them into values ​​within the range of [0,1]. The calculation expression is as follows: In the formula, X Norm This represents the standardized data of travel time and feature data; X chushi This represents the original travel time and feature data; min(X) and max(X) are the minimum and maximum values ​​in the original data, respectively. S2.4 Construct a travel time prediction dataset and divide it into training and test sets.

4. The method for guiding vehicle lane changing in highway tunnels based on a travel time prediction model according to claim 3, characterized in that: In step S4.2, the threshold value ΔT for the travel time difference is analyzed. threshold A multi-objective function, which evaluates the traffic efficiency after implementing the lane-changing strategy, is selected based on average travel time, average lost time, and average fuel consumption. The objective function is as follows: In the formula, β1, β2, and β3 represent the average travel time, respectively. Average Loss Time Average fuel consumption The correlation coefficients are β1 = β2 = β3 = 1.

5. A method for guiding vehicle lane changing in highway tunnels based on a travel time prediction model according to claim 4, characterized in that: In step S4.3, the optimal number of lane changes N change The calculation steps are as follows: Using average travel time Average Loss Time Average fuel consumption The objective function is used to analyze the optimal number of lane changes. By minimizing the objective function, the difference in travel time prediction ΔTT for each trip is obtained. pre The optimal number of lane changes; In the formula,

Images