A highway toll station emergency control method and system based on deep learning

Abstract

The present invention relates to the field of traffic management technology, and in particular to a method and system for emergency control of highway toll stations based on deep learning. The method includes obtaining vehicle data information, including obtaining historical vehicle data and real-time vehicle data of a highway section, preprocessing the obtained vehicle data, including cleaning and smoothing the obtained vehicle data, and normalizing the processed vehicle data; constructing a prediction model based on the preprocessed vehicle data, training the model according to the constructed prediction model, inputting the obtained real-time vehicle data into the trained prediction model, and obtaining the prediction result of the traffic volume in the future time; generating an emergency control strategy according to the prediction result, and the present invention can obtain vehicle data in real time and input the trained prediction model in time, so as to alleviate traffic congestion to the greatest extent and improve traffic efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic management, and in particular to a method and system for emergency control of highway toll stations based on deep learning. Background Art

[0002] As an important infrastructure of modern transportation, highways play an irreplaceable role in promoting regional economic exchanges and driving industrial development. As an important part of highways, the operating efficiency and management level of highway toll stations directly affect the smoothness and service quality of the entire transportation network. In the past, the management and control methods of highway toll stations have gone through a development process from manual operation to semi-automation, and then to a certain degree of automation.

[0003] However, traditional management and control methods have gradually exposed many shortcomings in dealing with increasingly complex and changeable traffic conditions. Traditional highway toll station management and control mainly rely on manual observation and simple equipment monitoring, and the accuracy and real-time performance of data collection are difficult to guarantee. In terms of data processing and analysis, traditional methods are usually based on simple statistical models and empirical formulas, which are difficult to cope with the complex nonlinear characteristics of traffic flow. For the prediction of traffic flow, the average method or linear regression method based on historical data is generally used. These methods cannot accurately capture the short-term fluctuations and long-term trends of traffic flow, especially in holidays and other conditions with heavy traffic. The prediction error is large, which brings great difficulties to the management of toll stations.

[0004] In addition, in terms of emergency management, traditional control methods are particularly weak. They cannot adjust charging strategies and traffic diversion measures in a timely and effective manner according to real-time traffic conditions, which leads to further aggravation of traffic congestion and may even cause large-scale traffic paralysis. Moreover, due to the lack of in-depth analysis and utilization of real-time traffic data, it is impossible to timely and accurately predict changes in traffic flow, and it is difficult to quickly generate effective emergency control strategies, resulting in the inability to achieve reasonable allocation and efficient utilization of traffic resources. This not only affects the traffic efficiency and service quality of highways, but also increases the risk of traffic accidents. At this stage, a highway toll station emergency control method and system based on deep learning is needed. Summary of the invention

[0005] In order to solve the problem that highway toll stations lack in-depth analysis and utilization of traffic data and cannot make emergency control in time, the present invention provides a highway toll station emergency control method and system based on deep learning. Through the improved Transformer model architecture and dynamic position coding, high-precision prediction of future traffic flow can be achieved.

[0006] In the first aspect, the present invention provides a method for emergency control of highway toll stations based on deep learning, which adopts the following technical solutions:

[0007] A highway toll station emergency control method based on deep learning, comprising:

[0008] Acquiring vehicle data information, including acquiring historical vehicle data and real-time vehicle data on expressways, wherein the vehicle data is specifically represented by vehicle speed data, flow data, location data and time data;

[0009] Preprocessing the acquired vehicle data;

[0010] Constructing a prediction model based on preprocessed vehicle data, including building a Transformer model architecture and introducing a position encoding of a dynamically adjusted combination of sine and cosine functions based on the Transformer model architecture;

[0011] Perform model training based on the constructed prediction model, including using the loss function to calculate the error between the predicted value and the actual value, and calculating the gradient of the loss function to the model parameters through the back propagation algorithm;

[0012] Input the acquired real-time vehicle data into the trained prediction model to obtain the prediction results of traffic flow in the future;

[0013] Generate emergency control strategies based on the prediction results, including using particle swarm optimization algorithm to screen multiple emergency control strategies.

[0014] Furthermore, the acquired vehicle data is preprocessed, including identifying outliers for each vehicle data according to preset outlier identification criteria, smoothing the remaining vehicle data using mean filtering, and selecting a standardization method based on the characteristics of each vehicle data, wherein a Z-score standardization operation is performed on the vehicle speed data, a Min-Max standardization operation is performed on the traffic data, and the position data is normalized to the [0,1] interval.

[0015] Furthermore, the construction of the prediction model based on the preprocessed vehicle data also includes adding residual connections to the position encoding Transformer model architecture to directly add shallow features to deep features, and finally using layer normalization to standardize the input of each layer in the Transformer.

[0016] Furthermore, the position encoding of the dynamically adjusted sine and cosine function combination introduced based on the Transformer model architecture includes defining time-related variables according to the time series of vehicle data, and constructing a dynamically adjusted position encoding formula using the variables, and the position encoding formula is:

[0017] ,

[0018] ,

[0019] in, Represented as a position index, Represented as a vector index, Expressed as model dimensions, Expressed as a time-varying scaling factor that adjusts the amplitude of the sine and cosine functions, It is expressed as a parameter that affects the frequency, so that the position encoding has different change characteristics at different times.

[0020] Furthermore, the model training is performed based on the constructed prediction model, including calculating the error between the predicted value and the actual value using the mean square error loss function, back-propagating the loss value along the hierarchical structure of the model through the back-propagation algorithm, calculating the gradient of each parameter to the loss, and updating the parameters of the model according to the stochastic gradient descent algorithm.

[0021] Furthermore, the loss value is back-propagated along the hierarchical structure of the model through a back-propagation algorithm, including a back-propagation mechanism based on an autoencoder, using an encoder to compress input data into a low-dimensional hidden layer representation, reconstructing the original input from the encoding through a decoder, defining a loss function and calculating gradients based on the loss value to update the parameters of the encoder and decoder, so as to gradually optimize the performance of the network and minimize the reconstruction error.

[0022] Furthermore, the particle swarm optimization algorithm is used to screen a variety of emergency control strategies, including representing each generated emergency control strategy as the position of a particle in the solution space, and the particle updates its position through multiple iterations according to its own historical optimal position and the optimal position of the entire group. The position update formula is:

[0023] ,

[0024] ,

[0025] in, and are the speeds of particle i at the tth and t+1th iterations in dimension d, and are the positions of particle i at the tth and t+1th iterations in dimension d, respectively. is the inertia weight, which controls the degree to which the particle maintains its original velocity. and is the learning factor, and is a random number in the range [0,1].

[0026] Second, a highway toll station emergency control system based on deep learning, including:

[0027] The data acquisition module is configured to: acquire vehicle data information, including acquiring historical vehicle data and real-time vehicle data of a highway section, wherein the vehicle data is specifically represented by vehicle speed data, flow data, location data and time data;

[0028] The preprocessing module is configured to: preprocess the acquired vehicle data;

[0029] A model module is configured to: construct a prediction model based on the preprocessed vehicle data, including constructing a Transformer model architecture, and introducing a position encoding of a dynamically adjusted combination of sine and cosine functions based on the Transformer model architecture;

[0030] The training module is configured to: perform model training according to the constructed prediction model, including calculating the error between the predicted value and the actual value by using the loss function, and calculating the gradient of the loss function to the model parameters by the back propagation algorithm;

[0031] The prediction module is configured to: input the acquired real-time vehicle data into the trained prediction model to obtain the prediction result of the traffic flow in the future;

[0032] The output module is configured to generate an emergency control strategy based on the prediction results, including screening a variety of emergency control strategies using a particle swarm optimization algorithm.

[0033] In a third aspect, the present invention provides a computer-readable storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, a method for emergency control of highway toll stations based on deep learning.

[0034] In a fourth aspect, the present invention provides a terminal device comprising a processor and a computer-readable storage medium, the processor being used to implement various instructions; the computer-readable storage medium being used to store multiple instructions, the instructions being suitable for being loaded by the processor and executing the method for emergency control of highway toll stations based on deep learning.

[0035] In summary, the present invention has the following beneficial technical effects:

[0036] 1. The present invention obtains comprehensive vehicle data, including multi-dimensional information such as speed, flow, location and time, and combines it with advanced Transformer model architecture and dynamic position coding to achieve high-precision prediction of future vehicle flow, which helps to predict traffic congestion and other conditions in advance and provide a certain reference basis for emergency control.

[0037] 2. The present invention introduces residual connection and layer normalization in model construction and training, which effectively solves the problems of gradient vanishing and gradient exploding in deep neural networks, enabling them to better adapt to different traffic scenarios and data distributions. It also adopts a back-propagation mechanism based on an automatic encoder to automatically learn and compress features during the back-propagation process, significantly improving the accuracy and efficiency of gradient calculation, accelerating the convergence speed of the model, and reducing the interference of redundant information on training.

[0038] 3. The present invention can obtain vehicle data in real time and input the trained prediction model in time, quickly generate the latest prediction results and corresponding emergency control strategies, and use the particle swarm optimization algorithm to screen the emergency control strategies. It can quickly find the optimal or near-optimal solution from many possible strategies, thereby maximally alleviating traffic congestion and improving traffic efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a schematic diagram of the overall process of a highway toll station emergency control method based on deep learning in an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0041] Example 1

[0042] Reference Figure 1 , a highway toll station emergency control method based on deep learning in this embodiment includes:

[0043] Acquiring vehicle data information, including acquiring historical vehicle data and real-time vehicle data on expressways, wherein the vehicle data is specifically represented by vehicle speed data, flow data, location data and time data;

[0044] Preprocessing the acquired vehicle data, including cleaning and smoothing the acquired vehicle data, and normalizing the processed vehicle data;

[0045] Constructing a prediction model based on preprocessed vehicle data, including building a Transformer model architecture and introducing a position encoding of a dynamically adjusted combination of sine and cosine functions based on the Transformer model architecture;

[0046] Perform model training based on the constructed prediction model, including using the loss function to calculate the error between the predicted value and the actual value, and calculating the gradient of the loss function to the model parameters through the back propagation algorithm;

[0047] Input the acquired real-time vehicle data into the trained prediction model to obtain the prediction results of traffic flow in the future;

[0048] Generate emergency control strategies based on the prediction results, including using particle swarm optimization algorithm to screen multiple emergency control strategies.

[0049] Specifically, a highway toll station emergency control method based on deep learning includes the following steps:

[0050] S1. Acquiring vehicle data information, including acquiring historical vehicle data and real-time vehicle data of a highway section, wherein the vehicle data is specifically represented by vehicle speed data, flow data, location data and time data;

[0051] Based on the traffic management system database, historical vehicle data and real-time vehicle data are obtained, specifically the vehicle speed, flow, and location information in different time periods and different road sections.

[0052] S2. Preprocessing the acquired vehicle data, including cleaning and smoothing the acquired vehicle data, and normalizing the processed vehicle data;

[0053] According to the preset outlier identification criteria, outlier identification is performed on each vehicle data. First, the mean (μ) and standard deviation (σ) of the speed data are calculated, and the data with speed values ​​outside the range of (μ-3σ, μ+3σ) are marked as outliers. At the same time, combined with the actual speed limit of the road, for example, if the road speed limit is 120 kilometers per hour, then data exceeding 180 kilometers per hour (with a 50% margin) is also regarded as an outlier. For these outliers, the average value of several adjacent normal speed values ​​is used to replace them. For the traffic flow data, the 5th percentile (P5) and 95th percentile (P95) of the traffic flow data are calculated, and the data with a speed less than P5 or greater than P95 are marked as outliers. 95 data is marked as an outlier. In addition, the upper and lower limits are set according to the historical traffic data and traffic capacity of the road section. For example, the maximum traffic of a certain section usually does not exceed 800 vehicles / hour, so data exceeding 1000 vehicles / hour may be abnormal. For outliers, the median is used for replacement. Finally, the rationality of the time format and time sequence is checked to ensure that the time data is in a unified format. Based on the acquired vehicle data, including vehicle speed data V={100,95,120,50,110,180,90,...}, traffic data F={50,80,100,30,120,...}, position data P={2 00,350,150,400,...} and the corresponding time data. For the processing of speed data, first calculate its mean μ. For example, assuming that the mean μ of the above speed data is calculated to be 100 and the standard deviation σ is 25, according to the set abnormal value standard, that is, the distance from the mean exceeds 3 times the standard deviation, for the remaining speed data, use the mean filter with a window size of 5 for smoothing. Taking the speed value of 95 as an example, the four data points around it are assumed to be 100, 110, 90 and 120, then the result after mean filtering is (95+100+110+90+120) / 5=103. For the flow data, first find its minimum value mi n_flow, assuming it is 30, and the maximum value max_flow, assuming it is 120. Next, the flow data is normalized by Min-Max. Taking the flow value of 80 as an example, the normalized result is (80-30) / (120-30)=0.5556. Through this standardization, flow data of different scales are unified to the same scale, which is convenient for subsequent comparison and analysis. For the location data, assuming its range is [100,500], each location data is normalized to the interval [0,1]. Taking the location value of 200 as an example, the normalized result is (200-100) / (500-100)=0.25. Such normalization makes the location data more comparable and universal in subsequent processing, associates data from different sources, associates vehicle information from the toll collection system with speed and location data detected by sensors to obtain a more comprehensive vehicle driving record, and finally sorts in chronological order: the collected data is arranged in chronological order so that the model can capture the characteristics of the time series. .

[0054] S3. Building a prediction model based on the preprocessed vehicle data, including building a Transformer model architecture, and introducing a position encoding of a dynamically adjusted combination of sine and cosine functions based on the Transformer model architecture;

[0055] The original Transformer architecture uses an absolute sinusoidal position encoding, which can be expressed as:

[0056] ,

[0057] ,

[0058] in, Represents the value of the posth position and the 2ith dimension in the position encoding matrix, pos represents the position index, i represents the vector index, and the dimension of the model. Expressed as model dimensions, the present invention is improved based on absolute sine position coding and is designed to dynamically adjust the parameters of sine and cosine functions to better reflect the time sequence information in the traffic flow data. The position coding formula is:

[0059] ,

[0060] ,

[0061] in, Represented as a position index, Represented as a vector index, Expressed as model dimensions, Expressed as a time-varying scaling factor that adjusts the amplitude of the sine and cosine functions, It is expressed as a parameter that affects the frequency, making the position encoding have different changing characteristics at different times. Specifically, The function of is to control the degree of expansion and contraction of the sine and cosine functions, dynamically adjust the amplitude of the function according to the change of time, so as to give different weights or importance to the position code at different time points. The value of can make the model pay more attention to or emphasize the location information of certain specific time points. This affects the frequency characteristics of the sine and cosine functions, which makes the position code have different changing rhythms or periods at different times. The value will lead to different periodic changes in the position code, so as to capture the information of different time intervals or patterns in the time series. and To determine, we conducted a time series autocorrelation analysis on historical traffic flow data and found that traffic flow showed a weekly periodicity. Assuming that the mean traffic flow in a week is , the standard deviation is ,but and as follows:

[0062] ,

[0063] ,

[0064] Where t represents the specific day of the week (1 represents Monday, and so on).

[0065] Then we analyze the variance of traffic flow in different time periods. If the variance is larger, it means the fluctuation is larger. Assume that the variance of traffic flow in different hours of the day is ,

[0066] ,

[0067] ,

[0068] Here, t represents the hour (1-24), and then the residual connection is introduced. The residual connection directly adds the shallow features to the deep features. Its main function is to enhance information transmission. In the neural network, as the number of layers increases, the problem of gradient vanishing or gradient exploding may occur, making it difficult for the model to learn the deep features. The residual connection allows the signal to be transmitted more directly from the shallow layer to the deep layer, which alleviates the problem of gradient vanishing to a certain extent.

[0069] Finally, the input of each layer is standardized. Layer normalization stabilizes the data distribution by standardizing the input of each layer, making the model more robust to different inputs.

[0070] S4. Performing model training based on the constructed prediction model, including calculating the error between the predicted value and the actual value using a loss function, and calculating the gradient of the loss function to the model parameters using a back propagation algorithm;

[0071] First, the mean square error loss function is used to measure the difference between the predicted value and the actual value. Assume that the actual value is y and the model's predicted value is , the calculation formula of the mean square error loss function is:

[0072] ,

[0073] Where n is the number of samples, Σ represents the sum of all samples. By calculating this loss value, the inaccuracy of the model prediction can be quantified. Next, the loss value is back-propagated along the hierarchy of the model through the back-propagation algorithm. The purpose of back-propagation is to calculate the gradient of each parameter with respect to the loss. Wherein, the loss value is back-propagated along the hierarchy of the model through the back-propagation algorithm, including a back-propagation mechanism based on an autoencoder, which consists of an encoder and a decoder. The encoder is responsible for compressing the input data into a low-dimensional hidden layer representation (encoding), while the decoder is responsible for reconstructing the original input from this encoding.

[0074] First, the input data is input into the encoder, which is a neural network composed of multiple layers of neurons. It processes the input data through a series of linear transformations and nonlinear activation functions, thereby mapping the high-dimensional input data to a low-dimensional encoding space. By limiting the size of the hidden layer, the autoencoder is forced to learn a compact representation of the input data that captures the key features of the input data. During the backpropagation process, the network automatically adjusts the parameters of the encoder so that the decoder can reconstruct the original input from the encoding as accurately as possible. The backpropagation algorithm calculates the gradient and updates the parameters in the encoder and decoder according to the value of the loss function to gradually optimize the performance of the network and minimize the reconstruction error. After calculating the gradient of the loss, the Stochastic Gradient Descent (SGD) algorithm is used to update the parameters of the model. By repeatedly repeating this process of calculating the loss, backpropagating the gradient, and updating the parameters, the model gradually learns the optimal parameters, thereby improving the accuracy of the prediction.

[0075] S5. Input the acquired real-time vehicle data into the trained prediction model to obtain the prediction result of the traffic flow in the future;

[0076] First, the model encodes and embeds the input real-time vehicle speed data, flow data, location data, and time data. Then, it uses the multi-head attention mechanism in the Transformer architecture to capture the long-distance dependencies and spatiotemporal correlations in the data, uses improved position encoding to understand the temporal order information in the data, and further processes and learns feature representations through a feedforward neural network. During the training process, the model learns how to predict future traffic flow based on historical data and current input. When making actual predictions, the latest real-time data is input into the trained model, and the model can output predictions for traffic flow in the future.

[0077] S6. Generate emergency control strategies based on the prediction results, including screening multiple emergency control strategies using a particle swarm optimization algorithm;

[0078] First, when an emergency control strategy is generated based on the predicted results of traffic flow, the control strategy includes the prediction of mild congestion, which can appropriately increase the number of manual toll lanes or open some ETC backup lanes. In moderate congestion, temporary toll points are added, and some exit lanes are changed to entrance lanes to increase traffic capacity. In severe congestion, all toll lanes are fully opened, including the backup lanes that are usually closed, and the speed limit requirements of ETC lanes are temporarily cancelled to speed up traffic. The particle swarm optimization algorithm is used to screen out the optimal emergency control strategy. In the particle swarm optimization algorithm, each emergency control strategy is represented as a position of a particle in the solution space, and each particle in the particle swarm represents a possible strategy combination. The particle swarm optimization algorithm is used to screen multiple emergency control strategies, including representing each generated emergency control strategy as the position of a particle in the solution space. The particle updates its position through multiple iterations based on its own historical optimal position and the optimal position of the entire group. The position update formula is:

[0079] ,

[0080] ,

[0081] in, and are the speeds of particle i at the tth and t+1th iterations in dimension d, and are the positions of particle i at the tth and t+1th iterations in dimension d, respectively. is the inertia weight, which controls the degree to which the particle maintains its original velocity. and is the learning factor, and is a random number in the range of [0,1]. For example, when there are multiple possible combinations of the number of toll lanes open and different toll standards as emergency control strategies, each particle in the particle swarm represents a specific combination. During the iteration process, if a particle finds that its current strategy performs well in calculating indicators such as traffic flow, congestion relief effect, and traffic efficiency, it will continue to adjust in this direction. Through continuous iteration and adjustment, the particle swarm can search for the optimal or near-optimal emergency control strategy in the solution space to cope with different traffic conditions and needs.

[0082] Example 2

[0083] The difference between this embodiment and embodiment 1 is that this embodiment provides a highway toll station emergency control system based on deep learning, including:

[0084] The data acquisition module is configured to: acquire vehicle data information, including acquiring historical vehicle data and real-time vehicle data of a highway section, wherein the vehicle data is specifically represented by vehicle speed data, flow data, location data and time data;

[0085] The preprocessing module is configured to: preprocess the acquired vehicle data;

[0086] A model module is configured to: construct a prediction model based on the preprocessed vehicle data, including constructing a Transformer model architecture, and introducing a position encoding of a dynamically adjusted combination of sine and cosine functions based on the Transformer model architecture;

[0087] The training module is configured to: perform model training according to the constructed prediction model, including calculating the error between the predicted value and the actual value by using the loss function, and calculating the gradient of the loss function to the model parameters by the back propagation algorithm;

[0088] The prediction module is configured to: input the acquired real-time vehicle data into the trained prediction model to obtain the prediction result of the traffic flow in the future;

[0089] The output module is configured to generate an emergency control strategy based on the prediction results, including screening a variety of emergency control strategies using a particle swarm optimization algorithm.

[0090] A computer-readable storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device for a method for emergency control of a highway toll station based on deep learning.

[0091] A terminal device includes a processor and a computer-readable storage medium, the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded and executed by the processor to implement the highway toll station emergency control method based on deep learning.

[0092] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A highway toll station emergency control method based on deep learning, characterized in that: include: Acquiring vehicle data information, including acquiring historical vehicle data and real-time vehicle data on expressways, wherein the vehicle data is specifically represented by vehicle speed data, flow data, location data and time data; Preprocessing the acquired vehicle data; The prediction model is constructed based on the preprocessed vehicle data, including constructing a Transformer model architecture, and introducing a position encoding of a dynamically adjusted combination of sine and cosine functions based on the Transformer model architecture, including defining time-related variables according to the time series of the vehicle data, and constructing a dynamically adjusted position encoding formula using the variables, wherein the position encoding formula is: , , in, Represented as a position index, Represented as a vector index, Expressed as model dimensions, Expressed as a time-varying scaling factor that adjusts the amplitude of the sine and cosine functions, It is expressed as a parameter that affects the frequency, so that the position encoding has different change characteristics at different times, and then based on data analysis and To determine, we first conduct a time series autocorrelation analysis on the historical traffic flow data and analyze the variance of traffic flow in different time periods. Suppose the variance of traffic flow at different hours in a day is ,but and It is expressed as: , , Where, t represents hours, t=1,2,3,…24; Perform model training based on the constructed prediction model, including using the loss function to calculate the error between the predicted value and the actual value, and calculating the gradient of the loss function to the model parameters through the back propagation algorithm; Input the acquired real-time vehicle data into the trained prediction model to obtain the prediction results of traffic flow in the future; Generate emergency control strategies based on the prediction results, including using particle swarm optimization algorithm to screen multiple emergency control strategies.

2. According to the method of emergency control of highway toll stations based on deep learning in claim 1, it is characterized in that: The obtained vehicle data is preprocessed, including identifying outliers for each vehicle data according to a preset outlier identification criterion, smoothing the remaining vehicle data using a mean filter, and selecting a standardization method according to the characteristics of each vehicle data, wherein a Z-score standardization operation is performed on the vehicle speed data, a Min-Max standardization operation is performed on the flow data, and the position data is normalized to the [0,1] interval.

3. According to the method of emergency control of highway toll stations based on deep learning in claim 2, it is characterized in that: The construction of the prediction model based on the preprocessed vehicle data also includes adding a residual connection to the position encoding Transformer model architecture to directly add shallow features to deep features, and finally using layer normalization to standardize the input of each layer in the Transformer.

4. According to the method of emergency control of highway toll stations based on deep learning in claim 3, it is characterized in that: The model training is performed based on the constructed prediction model, including using the mean square error loss function to calculate the error between the predicted value and the actual value, back-propagating the loss value along the hierarchical structure of the model through the back-propagation algorithm, calculating the gradient of each parameter to the loss, and updating the parameters of the model according to the stochastic gradient descent algorithm.

5. A highway toll station emergency control method based on deep learning according to claim 4, characterized in that: The loss value is back-propagated along the hierarchical structure of the model through a back-propagation algorithm, including a back-propagation mechanism based on an automatic encoder, using an encoder to compress input data into a low-dimensional hidden layer representation, reconstructing the original input from the encoding through a decoder, defining a loss function and calculating gradients based on the loss value to update the parameters of the encoder and decoder, so as to gradually optimize the performance of the network and minimize the reconstruction error.

6. A highway toll station emergency control method based on deep learning according to claim 5, characterized in that: The particle swarm optimization algorithm is used to screen a variety of emergency control strategies, including representing each generated emergency control strategy as the position of a particle in the solution space, and the particle updates its position through multiple iterations according to its own historical optimal position and the optimal position of the entire group. The position update formula is: , , in, and are the speeds of particle i at the tth and t+1th iterations in dimension d, and are the positions of particle i at the tth and t+1th iterations in dimension d, respectively. is the inertia weight, which controls the degree to which the particle maintains its original velocity. and is the learning factor, and is a random number in the range [0,1].

7. A highway toll station emergency control system based on deep learning, executing the method according to any one of claims 1 to 6, characterized in that: include: The data acquisition module is configured to: acquire vehicle data information, including acquiring historical vehicle data and real-time vehicle data of a highway section, wherein the vehicle data is specifically represented by vehicle speed data, flow data, location data and time data; The preprocessing module is configured to: preprocess the acquired vehicle data; A model module is configured to: construct a prediction model based on the preprocessed vehicle data, including constructing a Transformer model architecture, and introducing a position encoding of a dynamically adjusted combination of sine and cosine functions based on the Transformer model architecture; The training module is configured to: perform model training according to the constructed prediction model, including calculating the error between the predicted value and the actual value by using the loss function, and calculating the gradient of the loss function to the model parameters by the back propagation algorithm; The prediction module is configured to: input the acquired real-time vehicle data into the trained prediction model to obtain the prediction result of the traffic flow in the future; The output module is configured to generate an emergency control strategy based on the prediction results, including screening a variety of emergency control strategies using a particle swarm optimization algorithm.

8. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instruction is suitable for being loaded by a processor of a terminal device and executing a highway toll station emergency control method based on deep learning as described in any one of claims 1-6.

9. A terminal device, comprising a processor and a computer-readable storage medium, wherein the processor is used to implement each instruction; and the computer-readable storage medium is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing a highway toll station emergency control method based on deep learning as described in any one of claims 1-6.

Images