Expressway bottleneck section emergency lane management and control system

Abstract

The application discloses a kind of expressway bottleneck road section emergency lane management and control system, it is related to expressway emergency lane management and control technical field.The present application includes bottleneck road section entrance camera, network connection device, background processor, traffic guidance device;With bottleneck road section entrance camera acquisition video data, through network connection device transmission to background processor, utilize depth learning algorithm conversion into sub-vehicle traffic volume and speed, preprocessed data uses EMD-KNN-GRU combination model to carry out flow prediction, for congestion, construction, different scene of accident, according to service level and speed comprehensive judgment whether need to open emergency lane, and show current closed lane and open lane by traffic guidance device, show emergency lane closing information after road section recovers smooth.This application can effectively reduce expressway congestion time, avoid over long time to open emergency lane to cause the waste of emergency resource, investment cost is less, design is simple, and reliability is high.

Description

Technical Field

[0001] This invention belongs to the field of highway emergency lane control technology, and in particular relates to an emergency lane control system for highway bottleneck sections. Background Technology

[0002] In recent years, with the development of the social economy, highway traffic volume has continued to grow, and slow-moving traffic frequently occurs during holidays. To solve the persistent problems of highway traffic, it is essential to adjust traffic management and control strategies and plans in a timely manner to adapt to the ever-changing traffic demands, thereby improving the operational efficiency of the highway system and reducing social and economic losses.

[0003] Opening emergency lanes and temporarily using them as regular roads can effectively alleviate traffic congestion and improve the operational efficiency of the highway system. However, if opening emergency lanes still fails to quickly alleviate congestion, it will seriously affect their function as "lifelines." The question of how to effectively open emergency lanes without excessively compromising their emergency function remains unresolved. Summary of the Invention

[0004] This invention provides a highway bottleneck section emergency lane management system. Based on traffic volume data obtained through video recognition, the system uses an EMD-KNN-GRU combined model to predict future traffic volume and determine whether opening the emergency lane will cause prolonged congestion, affecting its emergency function. This achieves the goal of opening the emergency lane reasonably without excessively impacting its emergency function, thus making efficient use of road capacity while minimizing prolonged emergency lane occupancy. This invention rationally utilizes existing highway cameras and traffic guidance devices to increase the capacity of bottleneck sections, effectively reducing highway congestion time. It features low investment costs, simple design, and high reliability. The use of the EMD-KNN-GRU combined model overcomes the shortcomings of single models; the data obtained through EMD decomposition has better stability, improving the model's prediction accuracy. EMD decomposition, compared to common Fourier transforms and wavelet transforms, has adaptive characteristics, overcoming the limitations of Fourier transforms and wavelet transforms and improving the model's adaptability. In summary, this invention solves the problems in the background technology.

[0005] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0006] The present invention provides an emergency lane control system for bottleneck sections of highways, comprising a camera at the entrance of the bottleneck section, a network connection device, a background processor, and a traffic guidance device at the exit of the bottleneck section.

[0007] The bottleneck section entrance camera is installed at the entrance of the bottleneck section, including a bottleneck section entrance traffic guidance device and a bottleneck section exit traffic guidance device, used to acquire vehicle video information at the bottleneck section entrance;

[0008] The traffic guidance device is installed at the entrance and exit of the bottleneck section to indicate to vehicles the opening and closing of the emergency lane;

[0009] The output of the bottleneck section entrance camera is connected to the input of the network connection device, the output of the network connection device is connected to the input of the background processor, and the output of the background processor is connected to the traffic guidance device.

[0010] The background processor outputs emergency lane opening and closing information through the following steps:

[0011] S1. Acquisition of traffic flow and speed data by vehicle type: Using a fixed time period t0 as the detection cycle, deep learning-based video recognition is used to acquire and store traffic flow and speed data by vehicle type for bottleneck sections of highways.

[0012] S2. Calculate the ratio of traffic volume to baseline capacity: Determine the service level of the current bottleneck section based on the ratio of traffic volume to baseline capacity, and determine whether there is congestion based on the current vehicle speed;

[0013] S3. If the current bottleneck section is not congested, output the emergency lane closure information. If the current bottleneck section is congested, use the EMD-KNN-GRU combined model to predict the traffic volume for N detection cycles. Combine the number of lanes closed due to construction and accidents and the duration of lane closure to calculate the service level for N detection cycles after the emergency lane is opened, and determine whether the emergency lane can be restored to smooth traffic within N detection cycles after it is opened.

[0014] Specifically: If the current bottleneck section is not congested, output the emergency lane closure information; if the current bottleneck section is congested, normalize the traffic volume data and perform EMD decomposition. Input the high-frequency IMF component sequence obtained from the decomposition into the KNN model for prediction, and input the low-frequency IMF component sequence into the GRU model for prediction. After reconstructing the predicted components, obtain the predicted value; in congested scenarios, calculate the ratio of predicted traffic volume to baseline capacity after the emergency lane is opened, obtain the service level for N detection cycles after the emergency lane is opened, and determine whether the emergency lane can restore smooth traffic within N detection cycles; in construction and event scenarios, calculate the ratio of predicted traffic volume to baseline capacity after the construction or event lane is closed and the emergency lane is opened, obtain the service level for N detection cycles after the emergency lane is opened, and determine whether the emergency lane can restore smooth traffic within N detection cycles.

[0015] The normalization formula is: Z = (XX) min ) / (X max -Xmin ), where Z is the normalized value; X is the sample data; X max X represents the maximum value of the sample data. min The minimum value of the sample data;

[0016] The EMD decomposition is based on intrinsic mode functions, and the components must meet the following two conditions: the number of extreme points and the number of zero points of the processed signal in the time domain must be less than or equal to 1; the upper and lower envelopes of the processed signal must be symmetrical about the time axis. The specific steps are as follows:

[0017] P1. Finding extreme points: Finding the extreme points of the original signal X(t), including maximum and minimum values;

[0018] P2. Use cubic spline interpolation to connect all the maxima to form an upper envelope line e. max (t), connecting all local minimum points to form the lower envelope e min (t);

[0019] P3. Calculate the average envelope: The formula is: Where m1(t) is the average envelope, e max (t) is the upper envelope, e min (t) is the lower envelope. Subtracting the average envelope m1(t) from the original signal X(t) yields a new signal h1(t).

[0020] P4. Iteration: If h1(t) does not satisfy the two conditions of the intrinsic mode function, then replace X(t) with h1(t) and repeat the above three steps:

[0021] h 11 (t)=h1(t)-m 11 (t)

[0022] Iterate k times until h 1k (t) satisfies the two conditions of the intrinsic mode function;

[0023] h 1k If (t) is denoted as c1(t), then c1(t) is the first eigenmode function component of the original signal X(t), and the first eigenmode function component c1(t) is separated from the original signal to obtain the difference value r1(t);

[0024] r1(t) = X(t) - c1(t)

[0025] Then replace the original signal in step (1) with r1(t), and repeat steps P1, P2, and P3 to obtain the second, third, and nth intrinsic mode function components c2(t), c3(t), ..., c n (t), then according to r n (t)=rn-1 (t)-c n (t) Repeat the above steps and gradually separate the individual intrinsic mode function components of the original signal;

[0026] When c n The screening process terminates when (t) exhibits monotonic characteristics or is less than a preset value. Finally, the original signal can be represented as the sum of multiple intrinsic mode functions and residual functions, i.e.: c i (t) represents the i-th eigenmode function component of the original signal X(t), r n (t) is the last remaining signal, called the residual function, r n (t) represents the stationary trend of the signal, while the eigenmode function components c of each order i (t) sequentially contains different frequency band components of signal X(t) from high to low;

[0027] The KNN model works by selecting the q periods of data preceding the prediction period as the state vector, matching the state vector against historical data to obtain the K sets of historical data closest to the prediction period, and using the arithmetic mean of these K sets of historical data as the predicted value. The distance metric reflects the approximation between the historical and current state vectors, using Euclidean distance as the distance metric. The expression for the Euclidean distance in n-dimensional space is: Where L(x) i ,x j ) is x i With x j The Euclidean distance between them, x i Let x be the historical state vector. j This is the current state vector;

[0028] The GRU model improves upon the structure of a recurrent neural network (RNN) to address issues like vanishing and exploding gradients during long-sequence training. Similar to LSTM, GRU uses a gating mechanism to control input and memory information at the current time step to make predictions. With only two control gates, GRU reduces computational complexity and improves training efficiency. Each unit in the GRU model includes an update gate z. t (t is the time step), reset gate r t Candidate vector h t ′、New state vector h t ;x t Given the input vector at time step t, a threshold is formed by matrix concatenation and an activation function; the gate z is then updated. t The degree to which past memory information is retained in the current state is determined by the expression z.t =σ(W (z) x t +U (z) h t-1 ), where: z t For the update gate, σ is the Sigmoid activation function, and W... (z) U (z) To update the training parameter matrix in the gate, x t h is the input of the hidden layer at time t. t-1 Output at time t-1; reset gate r t It determines how previously memorized state information and new input information are combined, expressed as r. t =σ(W (r) x t +U (r) h t-1 ), where: r t For the reset gate, σ is the Sigmoid activation function, and W... (r) U (r) To reset the training parameter matrix in the gate, x t h is the input of the hidden layer at time t. t-1 This is the output at time t-1; the new memory content will store past related information by resetting the gate output, expressed as h. t =tanh(Wx) t +r t ⊙Uh t-1 ), where: h t ′ is the input x t The output h from the previous time step t-1 In summary, tanh is the hyperbolic tangent activation function, and W and U are used to calculate h. t The training parameter matrix during the process, x t Let r be the input of the hidden layer at time t. t To reset the door, h t-1 The output at time t-1 is represented by ⊙, which represents the Hadamard product operation; the new state vector h is... t The information from the current cell will be retained and passed to the next cell; the expression is h. t =z t ⊙h t-1 +(1-z t )⊙h t ′, where: h t Let z be the output of the current layer at time t. t For the update gate, ⊙ represents the Hadamard product operation, h t-1 h is the output at time t-1. t ′ is the input x t The output h from the previous time stept-1 In summary, the training of the GRU model consists of two processes: forward propagation and backward error propagation. The forward propagation process calculates the output value of the GRU unit, while the backward error propagation process calculates the error between the output value and the actual value. Following the gradient descent method, the connection weights of each layer of neurons are updated sequentially by taking the derivative in the direction of decreasing error gradient. The two processes of forward propagation and backward error propagation alternate continuously until the error is minimized or the maximum number of iterations is reached, at which point the iteration process ends.

[0029] The output result is inversely normalized using the formula: X = Z(X max -X min )+X min Where Z is the value before denormalization, and X is the value after denormalization. max X represents the maximum value of the sample data. min The minimum value of the sample data;

[0030] Output predicted value Y i1 Y i2 The final prediction result is obtained, namely: Y = Y i1 +Y i2 , where Y i1 Y is the predicted value of the KNN model. i2 GRU model predicted values

[0031] S4. If it is determined that the emergency lane is still congested after N detection cycles after it is opened, output the emergency lane closure information; if it is determined that the emergency lane is clear after N detection cycles after it is opened, output the emergency lane opening information.

[0032] S5. Repeat steps S1 through S4.

[0033] Furthermore, the bottleneck section entrance camera is a 4-megapixel network camera installed at a height of 6-8 meters.

[0034] Furthermore, the network connection device includes routers, switches, cables, wireless access points, and network interface cards to enable communication between the entrance camera of the bottleneck section and the back-end processor.

[0035] Furthermore, the traffic guidance device may employ variable message signs, broadcasts, traffic guidance screens, public information service platforms, in-vehicle navigation systems, or mobile apps.

[0036] Furthermore, step S1 specifically includes:

[0037] Within the field of view, one or more detection regions, i.e., virtual loops, are defined according to lanes. The virtual loop vehicle detection module consists of background reconstruction, shadow removal, vehicle detection, and traffic parameter measurement. Through background reconstruction, the current frame image is segmented into background and non-background regions according to certain principles. The non-background region includes moving vehicles and the shadows cast by vehicles. Through shadow detection, the image is segmented into background, foreground, and shadow. Shadow removal is used to obtain the foreground segmentation image. Then, based on the foreground segmentation image, it can be determined whether a vehicle has passed through the virtual loop according to certain principles. Based on the virtual loop, the traffic parameter detection module is used to statistically analyze traffic parameters such as traffic volume and vehicle speed data.

[0038] Furthermore, step S2 specifically includes:

[0039] The system preprocesses traffic flow and speed data by vehicle type, removing abnormal and duplicate data and filling in missing data. It calculates the ratio of traffic volume to baseline capacity and assesses whether the service level of the current bottleneck section is at least Level 3 according to the highway engineering technical standards. If the current service level is below Level 3 and the current speed is below the threshold, the bottleneck section is judged to be congested. If the current service level is at least Level 3 or the current speed is at least the threshold, the bottleneck section is judged to be unobstructed.

[0040] The present invention has the following advantages over the prior art:

[0041] (1) Based on traffic volume data obtained from video recognition, this invention uses the EMD-KNN-GRU combined model to predict future traffic volume and determine whether opening the emergency lane will cause long-term congestion that cannot be dissipated, affecting the emergency function of the emergency lane. This achieves the effect of opening the emergency lane reasonably without excessively affecting its emergency function; it makes reasonable use of road capacity while minimizing the long-term occupation of the emergency lane.

[0042] (2) This invention makes reasonable use of existing highway cameras and traffic guidance devices to increase the traffic capacity of bottleneck sections, which can effectively reduce highway congestion time, with low investment cost, simple design and high reliability.

[0043] (3) This invention uses the EMD-KNN-GRU combined model to solve the defects of the single model. The data obtained by EMD decomposition has better stationarity and can improve the prediction accuracy of the model. EMD decomposition has the characteristic of adaptability compared with common Fourier transform and wavelet transform, which overcomes the limitations of Fourier transform and wavelet transform and improves the adaptability of the model.

[0044] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is an installation structure diagram of an emergency lane control system for bottleneck sections of highways according to the present invention;

[0047] Figure 2 for Figure 1 Connection diagram of each component;

[0048] Figure 3 This is a flowchart of the emergency lane control system for bottleneck sections of highways according to the present invention;

[0049] Figure 4 This is a schematic diagram of the combined model in the system of the present invention;

[0050] Figure 5 This is a schematic diagram of the GRU model in the system of this invention;

[0051] The attached diagram lists the components represented by each number as follows:

[0052] 1-Emergency lane, 2-Traffic guidance device at the entrance of bottleneck section, 3-Camera at the entrance of bottleneck section, 4-Traffic guidance device at the exit of bottleneck section. Detailed Implementation

[0053] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Please see Figure 1-5 As shown, the present invention provides a highway bottleneck section emergency lane control system, including a bottleneck section entrance camera 3, a network connection device, a background processor, and a traffic guidance device at the bottleneck section exit.

[0055] The bottleneck section entrance camera 3 is installed at the entrance of the bottleneck section to obtain video information of vehicles at the entrance of the bottleneck section;

[0056] Traffic guidance devices are installed at the entrance and exit of the bottleneck section, including a bottleneck section entrance traffic guidance device 2 and a bottleneck section exit traffic guidance device 4, which are used to indicate the opening and closing information of the emergency lane 1 to vehicles;

[0057] The output of the bottleneck section entrance camera 3 is connected to the input of the network connection device, the output of the network connection device is connected to the input of the background processor, and the output of the background processor is connected to the traffic guidance device.

[0058] The background processor outputs the opening and closing information of emergency lane 1 through the following steps:

[0059] S1. Acquisition of traffic flow and speed data by vehicle type: Using a fixed time period t0 as the detection cycle, deep learning-based video recognition is used to acquire and store traffic flow and speed data by vehicle type for bottleneck sections of highways.

[0060] S2. Calculate the ratio of traffic volume to baseline capacity: Determine the service level of the current bottleneck section based on the ratio of traffic volume to baseline capacity, and determine whether there is congestion based on the current vehicle speed;

[0061] S3. If the current bottleneck section is not congested, output the closure information of emergency lane 1. If the current bottleneck section is congested, use the EMD-KNN-GRU combined model to predict the traffic volume for N detection cycles. Combine the number of lanes closed due to construction and accidents and the duration of lane closure to calculate the service level of emergency lane 1 after it is opened for N detection cycles, and determine whether emergency lane 1 can be restored to smooth traffic within N detection cycles after it is opened.

[0062] Specifically: If the current bottleneck section is not congested, output the closure information of emergency lane 1; if the current bottleneck section is congested, normalize the traffic volume data and perform EMD decomposition. Input the high-frequency IMF component sequence obtained from the decomposition into the KNN model for prediction, and input the low-frequency IMF component sequence into the GRU model for prediction. After the predicted components are reconstructed, the predicted value is obtained. In congested scenarios, calculate the ratio of predicted traffic volume to baseline capacity after emergency lane 1 is opened, obtain the service level of emergency lane 1 after N detection cycles, and determine whether emergency lane 1 can restore smooth traffic within N detection cycles. In construction and event scenarios, calculate the ratio of predicted traffic volume to baseline capacity after the construction or event lane is closed and emergency lane 1 is opened, obtain the service level of emergency lane 1 after N detection cycles, and determine whether emergency lane 1 can restore smooth traffic within N detection cycles.

[0063] The normalization formula is: Z = (XX) min ) / (X max -X min ), where Z is the normalized value; X is the sample data; X max X represents the maximum value of the sample data. min The minimum value of the sample data;

[0064] The EMD is decomposed into Intrinsic Mode Functions (IMFs), and the components must meet the following two conditions: the number of extreme points and the number of zero points of the processed signal in the time domain must be less than or equal to 1; the upper and lower envelopes of the processed signal must be symmetrical about the time axis. The specific steps are as follows:

[0065] P1. Finding extreme points: Finding the extreme points of the original signal X(t), including maximum and minimum values;

[0066] P2. Use cubic spline interpolation to connect all the maxima to form an upper envelope line e. max (t), connecting all local minimum points to form the lower envelope e min (t);

[0067] P3. Calculate the average envelope: The formula is: Where m1(t) is the average envelope, e max (t) is the upper envelope, e min (t) is the lower envelope. Subtracting the average envelope m1(t) from the original signal X(t) yields a new signal h1(t).

[0068] P4. Iteration: If h1(t) does not satisfy the two conditions of the intrinsic mode function, then replace X(t) with h1(t) and repeat the above three steps:

[0069] h 11 (t)=h1(t)-m 11 (t)

[0070] Iterate k times until h 1k (t) satisfies the two conditions of the intrinsic mode function;

[0071] h 1k If (t) is denoted as c1(t), then c1(t) is the first eigenmode function component of the original signal X(t), and the first eigenmode function component c1(t) is separated from the original signal to obtain the difference value r1(t);

[0072] r1(t) = X(t) - c1(t)

[0073] Then replace the original signal in step (1) with r1(t), and repeat steps P1, P2, and P3 to obtain the second, third, and nth intrinsic mode function components c2(t), c3(t), ..., c n (t), then according to r n (t)=r n-1 (t)-c n (t) Repeat the above steps and gradually separate the individual intrinsic mode function components of the original signal;

[0074] When c n The screening process terminates when (t) exhibits monotonic characteristics or is less than a preset value. Finally, the original signal can be represented as the sum of multiple intrinsic mode functions and residual functions, i.e.: c i (t) represents the i-th eigenmode function component of the original signal X(t), r n (t) is the last remaining signal, called the residual function, r n (t) represents the stationary trend of the signal, while the eigenmode function components c of each order i (t) sequentially contains different frequency band components of signal X(t) from high to low;

[0075] The KNN model works by selecting the q periods of data preceding the prediction period as the state vector, matching the state vector against historical data to obtain the K sets of historical data closest to the prediction period, and using the arithmetic mean of these K sets of historical data as the predicted value. The distance metric reflects the approximation between the historical and current state vectors, using Euclidean distance as the distance metric. The expression for the Euclidean distance in n-dimensional space is: Where L(x) i ,x j ) is x i With x j The Euclidean distance between them, x i Let x be the historical state vector. j This is the current state vector;

[0076] The GRU model improves upon the structure of a recurrent neural network (RNN) to address issues like vanishing and exploding gradients during long-sequence training. Similar to LSTM, GRU uses a gating mechanism to control input and memory information at the current time step to make predictions. With only two control gates, GRU reduces computational complexity and improves training efficiency. Each unit in the GRU model includes an update gate z. t (t is the time step), reset gate r t Candidate vector h t ′、New state vector h t ;x t Given the input vector at time step t, a threshold is formed by matrix concatenation and an activation function; the gate z is then updated. t The degree to which past memory information is retained in the current state is determined by the expression z. t =σ(W (z) x t +U (z) h t-1), where: z t For the update gate, σ is the Sigmoid activation function, and W... (z) U (z) To update the training parameter matrix in the gate, x t h is the input of the hidden layer at time t. t-1 Output at time t-1; reset gate r t It determines how previously memorized state information and new input information are combined, expressed as r. t =σ(W (r) x t +U (r) h t-1 ), where: r t For the reset gate, σ is the Sigmoid activation function, and W... (r) U (r) To reset the training parameter matrix in the gate, x t h is the input of the hidden layer at time t. t-1 This is the output at time t-1; the new memory content will store past related information by resetting the gate output, expressed as h. t =tanh(Wx) t +r t ⊙Uh t-1 ), where: h t ′ is the input x t The output h from the previous time step t-1 In summary, tanh is the hyperbolic tangent activation function, and W and U are used to calculate h. t The training parameter matrix during the process, x t Let r be the input of the hidden layer at time t. t To reset the door, h t-1 The output at time t-1 is represented by ⊙, which represents the Hadamard product operation; the new state vector h is... t The information from the current cell will be retained and passed to the next cell; the expression is h. t =z t ⊙h t-1 +(1-z t )⊙h t ′, where: h t Let z be the output of the current layer at time t. t For the update gate, ⊙ represents the Hadamard product operation, h t-1 h is the output at time t-1. t ′ is the input x t The output h from the previous time step t-1In summary, the training of the GRU model consists of two processes: forward propagation and backward error propagation. The forward propagation process calculates the output value of the GRU unit, while the backward error propagation process calculates the error between the output value and the actual value. Following the gradient descent method, the connection weights of each layer of neurons are updated sequentially by taking the derivative in the direction of decreasing error gradient. The two processes of forward propagation and backward error propagation alternate continuously until the error is minimized or the maximum number of iterations is reached, at which point the iteration process ends.

[0077] The output result is inversely normalized using the formula: X = Z(X max -X min )+X min Where Z is the value before denormalization, and X is the value after denormalization. max X represents the maximum value of the sample data. min The minimum value of the sample data;

[0078] Output predicted value Y i1 Y i2 The final prediction result is obtained, namely: Y = Y i1 +Y i2 , where Y i1 Y is the predicted value of the KNN model. i2 GRU model predicted values

[0079] S4. If it is determined that the emergency lane 1 is still congested after N detection cycles, output the closure information of the emergency lane (1). If it is determined that the emergency lane 1 is clear after N detection cycles, output the opening information of the emergency lane 1.

[0080] S5. Repeat steps S1 through S4.

[0081] Among them, the bottleneck section entrance camera 3 is a 4-megapixel network camera installed at a height of 6-8 meters.

[0082] Among them, the network connection device includes routers, switches, cables, wireless access points, and network interface cards to enable communication between the bottleneck section entrance camera 3 and the back-end processor.

[0083] Traffic guidance devices can be implemented using variable message signs, broadcasts, traffic guidance screens, public information service platforms, in-vehicle navigation systems, or mobile apps.

[0084] Specifically, step S1 includes:

[0085] Within the field of view, one or more detection regions, i.e., virtual loops, are defined according to lanes. The virtual loop vehicle detection module consists of background reconstruction, shadow removal, vehicle detection, and traffic parameter measurement. Through background reconstruction, the current frame image is segmented into background and non-background regions according to certain principles. The non-background region includes moving vehicles and the shadows cast by vehicles. Through shadow detection, the image is segmented into background, foreground, and shadow. Shadow removal is used to obtain the foreground segmentation image. Then, based on the foreground segmentation image, it can be determined whether a vehicle has passed through the virtual loop according to certain principles. Based on the virtual loop, the traffic parameter detection module is used to statistically analyze traffic parameters such as traffic volume and vehicle speed data.

[0086] Specifically, step S2 includes:

[0087] The system preprocesses traffic flow and speed data by vehicle type, removing abnormal and duplicate data and filling in missing data. It calculates the ratio of traffic volume to baseline capacity and assesses whether the service level of the current bottleneck section is at least Level 3 according to the highway engineering technical standards. If the current service level is below Level 3 and the current speed is below the threshold, the bottleneck section is judged to be congested. If the current service level is at least Level 3 or the current speed is at least the threshold, the bottleneck section is judged to be unobstructed.

[0088] This invention, based on traffic volume data obtained through video recognition, uses an EMD-KNN-GRU combined model to predict future traffic volume and determine whether opening emergency lanes will cause prolonged congestion and affect their emergency function. This achieves the goal of opening emergency lanes reasonably without excessively impacting their emergency function, thus making efficient use of road capacity while minimizing the prolonged occupation of emergency lanes. This invention also rationally utilizes existing highway cameras and traffic guidance devices to increase capacity in bottleneck sections, effectively reducing highway congestion time. It is characterized by low investment costs, simple design, and high reliability.

[0089] This invention uses an EMD-KNN-GRU combined model to address the shortcomings of single models. The data obtained through EMD decomposition exhibits better stationarity, thereby improving the model's prediction accuracy. Compared to common Fourier transforms and wavelet transforms, EMD decomposition is adaptive, overcoming the limitations of Fourier transforms and wavelet transforms and improving the model's adaptability.

[0090] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A highway bottleneck section emergency lane control system, characterized in that, Includes a bottleneck section entrance camera (3), a network connection device, a background processor, and a traffic guidance device at the bottleneck section exit; The bottleneck section entrance camera (3) is installed at the entrance of the bottleneck section to obtain vehicle video information at the entrance of the bottleneck section; The traffic guidance device is installed at the entrance and exit of the bottleneck section, including a bottleneck section entrance traffic guidance device (2) and a bottleneck section exit traffic guidance device (4), which are used to indicate the opening and closing information of the emergency lane (1) for vehicles; The output end of the bottleneck section entrance camera (3) is connected to the input end of the network connection device, the output end of the network connection device is connected to the input end of the background processor, and the output end of the background processor is connected to the traffic guidance device. The background processor outputs the opening and closing information of the emergency lane (1) through the following steps: S1. Acquisition of traffic flow and speed data by vehicle type: Using a fixed time period t0 as the detection cycle, deep learning-based video recognition is used to acquire and store traffic flow and speed data by vehicle type for bottleneck sections of highways. S2. Calculate the ratio of traffic volume to baseline capacity: Determine the service level of the current bottleneck section based on the ratio of traffic volume to baseline capacity, and determine whether there is congestion based on the current vehicle speed; S3. If the current bottleneck section is not congested, output the closure information of the emergency lane (1). If the current bottleneck section is congested, use the EMD-KNN-GRU combined model to predict the traffic volume of N detection cycles. Combine the number of lanes closed due to construction and accidents and the duration of lane closure to calculate the service level of the emergency lane (1) after it is opened for N detection cycles. Determine whether the emergency lane (1) can be restored to smooth traffic within N detection cycles after it is opened. Specifically: if the current bottleneck section is not congested, output the closure information of the emergency lane (1); if the current bottleneck section is congested, normalize the traffic volume data and perform EMD decomposition, input the high-frequency IMF component sequence obtained by decomposition into the KNN model for prediction, input the low-frequency IMF component sequence into the GRU model for prediction, and obtain the predicted value after the predicted components are reconstructed. In congestion scenarios, the ratio of predicted traffic volume to baseline capacity after emergency lane (1) is opened is calculated to obtain the service level of emergency lane (1) for N detection cycles after it is opened, and it is determined whether emergency lane (1) can be restored to smooth traffic within N detection cycles after it is opened; in construction and event scenarios, the ratio of predicted traffic volume to baseline capacity after emergency lane (1) is opened is calculated to obtain the service level of emergency lane (1) for N detection cycles after it is opened, and it is determined whether emergency lane (1) can be restored to smooth traffic within N detection cycles after it is opened; The normalization formula is: Z = (XX) min ) / (X max -X min ), where Z is the normalized value; X is the sample data; X max X represents the maximum value of the sample data. min The minimum value of the sample data; The EMD decomposition is based on intrinsic mode functions, and the components must meet the following two conditions: the number of extreme points and the number of zero points of the processed signal in the time domain must be less than or equal to 1; the upper and lower envelopes of the processed signal must be symmetrical about the time axis. The specific steps are as follows: P1. Finding extreme points: Finding the extreme points of the original signal X(t), including maximum and minimum values; P2. Use cubic spline interpolation to connect all the maxima to form an upper envelope line e. max (t), connecting all local minimum points to form the lower envelope e min (t); P3. Calculate the average envelope: The formula is: Where m1(t) is the average envelope, e max (t) is the upper envelope, e min (t) is the lower envelope. Subtracting the average envelope m1(t) from the original signal X(t) yields a new signal h1(t). P4. Iteration: If h1(t) does not satisfy the two conditions of the intrinsic mode function, then replace X(t) with h1(t) and repeat the above three steps: h 11 (t)=h1(t)-m 11 (t) Iterate k times until h 1k (t) satisfies the two conditions of the intrinsic mode function; h 1k If (t) is denoted as c1(t), then c1(t) is the first eigenmode function component of the original signal X(t), and the first eigenmode function component c1(t) is separated from the original signal to obtain the difference value r1(t); r1(t) = X(t) - c1(t) Then replace the original signal in step (1) with r1(t), and repeat steps P1, P2, and P3 to obtain the second, third, and nth intrinsic mode function components c2(t), c3(t), ..., c n (t), then according to r n (t)=r n-1 (t)-c n (t) Repeat the above steps and gradually separate the individual intrinsic mode function components of the original signal; When c n The screening process terminates when (t) exhibits monotonic characteristics or is less than a preset value. Finally, the original signal can be represented as the sum of multiple intrinsic mode functions and residual functions, i.e.: c i (t) represents the i-th eigenmode function component of the original signal X(t), r n (t) is the last remaining signal, called the residual function, r n (t) represents the stationary trend of the signal, while the eigenmode function components c of each order i (t) sequentially contains different frequency band components of signal X(t) from high to low; The KNN model works by selecting the q periods of data preceding the prediction period as the state vector, matching the state vector against historical data to obtain the K sets of historical data closest to the prediction period, and using the arithmetic mean of these K sets of historical data as the predicted value. The distance metric reflects the approximation between the historical and current state vectors, using Euclidean distance as the distance metric. The expression for the Euclidean distance in n-dimensional space is: Where L(x) i ,x j ) is x i With x j The Euclidean distance between them, x i Let x be the historical state vector. j This is the current state vector; The GRU model improves upon the structure of a recurrent neural network (RNN) to address issues such as vanishing and exploding gradients during long training sequences. Within each unit of the GRU model, there is an update gate z. t (t is the time step), reset gate r t Candidate vector h t ′、New state vector h t ;x t Given the input vector at time step t, a threshold is formed by matrix concatenation and an activation function; the gate z is then updated. t The degree to which past memory information is retained in the current state is determined by the expression z. t =σ(W (z) x t +U (z) h t-1 ), where: z t For the update gate, σ is the Sigmoid activation function, and W... (z) U (z) To update the training parameter matrix in the gate, x t h is the input of the hidden layer at time t. t-1 Output at time t-1; reset gate r t It determines how previously memorized state information and new input information are combined, expressed as r. t =σ(W (r) x t +U (r) h t-1 ), where: r t For the reset gate, σ is the Sigmoid activation function, and W... (r) U (r) To reset the training parameter matrix in the gate, x t h is the input of the hidden layer at time t. t-1 This is the output at time t-1; the new memory content will store past related information by resetting the gate output, expressed as h. t =tanh(Wx) t +r t ⊙Uh t-1 ), where: h t ′ is the input x t The output h from the previous time step t-1 In summary, tanh is the hyperbolic tangent activation function, and W and U are used to calculate h. t The training parameter matrix during the process, x t Let r be the input of the hidden layer at time t. t To reset the door, h t-1 The output at time t-1 is represented by ⊙, which represents the Hadamard product operation; the new state vector h is... t The information from the current cell will be retained and passed to the next cell; the expression is h. t =z t ⊙h t-1 +(1-z t )⊙h t ′, where: h t Let z be the output of the current layer at time t. t For the update gate, ⊙ represents the Hadamard product operation, h t-1 h is the output at time t-1. t ′ is the input x t The output h from the previous time step t-1 In summary, the training of the GRU model consists of two processes: forward propagation and backward error propagation. The forward propagation process calculates the output value of the GRU unit, while the backward error propagation process calculates the error between the output value and the actual value. Following the gradient descent method, the connection weights of each layer of neurons are updated sequentially by taking the derivative in the direction of decreasing error gradient. The two processes of forward propagation and backward error propagation alternate continuously until the error is minimized or the maximum number of iterations is reached, at which point the iteration process ends. The output result is inversely normalized using the formula: X = Z(X max -X min )+X min Where Z is the value before denormalization, and X is the value after denormalization. max X represents the maximum value of the sample data. min The minimum value of the sample data; Output predicted value Y i1 Y i2 The final prediction result is obtained, namely: Y = Y i1 +Y i2 , where Y i1 Y is the predicted value of the KNN model. i2 These are the predicted values ​​from the GRU model; S4. If it is determined that the emergency lane (1) is still congested after N detection cycles after it is opened, output the closure information of the emergency lane (1). If it is determined that the emergency lane (1) is clear after N detection cycles after it is opened, output the opening information of the emergency lane (1). S5. Repeat steps S1 through S4.

2. The emergency lane control system for bottleneck sections of highways according to claim 1, characterized in that, The bottleneck section entrance camera (3) is a 4-megapixel network camera installed at a height of 6-8 meters.

3. The emergency lane control system for bottleneck sections of highways according to claim 1, characterized in that, The network connection device consists of a router, switch, cable, wireless access point, and network interface card, which enable communication between the entrance camera (3) of the bottleneck section and the background processor.

4. The emergency lane control system for bottleneck sections of highways according to claim 1, characterized in that, The traffic guidance device can be a variable message sign, broadcast, traffic guidance screen, public information service platform, vehicle navigation system or mobile APP.

5. The emergency lane control system for bottleneck sections of highways according to claim 1, characterized in that, The S1 step specifically includes: Within the field of view, one or more detection regions, i.e., virtual loops, are defined according to lanes. The virtual loop vehicle detection module consists of background reconstruction, shadow removal, vehicle detection, and traffic parameter measurement. Through background reconstruction, the current frame image is segmented into background and non-background regions according to certain principles. The non-background region includes moving vehicles and the shadows cast by vehicles. Through shadow detection, the image is segmented into background, foreground, and shadow. Shadow removal is used to obtain the foreground segmentation image. Then, based on the foreground segmentation image, it can be determined whether a vehicle has passed through the virtual loop according to certain principles. Based on the virtual loop, the traffic parameter detection module is used to statistically analyze traffic parameters such as traffic volume and vehicle speed data.

6. The emergency lane control system for bottleneck sections of highways according to claim 1, characterized in that, The S2 step specifically includes: The system preprocesses traffic flow and speed data by vehicle type, removing abnormal and duplicate data and filling in missing data. It calculates the ratio of traffic volume to baseline capacity and assesses whether the service level of the current bottleneck section is at least Level 3 according to the highway engineering technical standards. If the current service level is below Level 3 and the current speed is below the threshold, the bottleneck section is judged to be congested. If the current service level is at least Level 3 or the current speed is at least the threshold, the bottleneck section is judged to be unobstructed.

Images