Traffic flow induction monitoring method and system based on convolution denoising self-coding

Abstract

The invention discloses a traffic flow induction monitoring method and system based on convolution denoising self-coding, and relates to the technical field of traffic control, and the method comprises the steps: generating a fused space-time traffic data set; extracting vehicle-mounted and navigation data from the space-time traffic data set, and preliminarily dividing driving vehicles into vehicles with known navigation information and vehicles with unknown navigation information; processing the space-time traffic data of the vehicle with unknown navigation information by using a convolution denoising auto-encoder, a space-time convolution neural network and a long short-term memory network to generate a traffic flow prediction model; training and optimizing the traffic flow prediction model by using a joint loss function; according to the space-time traffic data of the vehicles with known navigation information and the vehicles with unknown navigation information, calculating the traffic index of the next road section, and if the traffic index of the next road section is greater than a preset traffic threshold, guiding the vehicles to switch to recommended alternative paths, providing optimal driving path suggestions, and realizing guidance and optimization of traffic flow.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic control, and in particular to a traffic flow induction monitoring method and system based on convolutional denoising autoencoding. Background Art

[0002] With the acceleration of urbanization and the continuous growth of traffic demand, traffic congestion has become increasingly serious, which has had a serious impact on people's travel efficiency and urban operation efficiency. In order to effectively alleviate traffic congestion and improve road capacity, traffic flow induction systems have emerged. Existing traffic flow induction systems usually use traditional data processing and prediction methods, such as time series analysis and machine learning algorithms. However, these methods often have problems such as low computational efficiency and insufficient prediction accuracy when processing large-scale, high-dimensional traffic data. At the same time, these methods are difficult to effectively capture the spatiotemporal characteristics of traffic data, that is, the interrelationships and dependencies of traffic data in space and time.

[0003] In recent years, the development of deep learning technology has provided new solutions for traffic flow induction systems, especially deep learning models such as convolutional neural networks (CNN), recurrent neural networks (RNN) and their variants (such as long short-term memory networks (LSTM)). They have achieved remarkable results in image recognition, speech recognition, natural language processing and other fields. These models have powerful feature extraction and pattern recognition capabilities, can process large-scale, high-dimensional data, and capture complex patterns and correlations in the data.

[0004] However, the existing traffic flow induction systems based on deep learning still have some shortcomings. For example, in the data preprocessing stage, the existing systems often use simple filtering and normalization methods, which makes it difficult to effectively remove noise and outliers in the data. In the prediction stage, the existing systems usually only consider the traffic data of all vehicles, while ignoring the mutual influence between vehicles with known navigation information and vehicles with unknown navigation information, resulting in inaccurate and incomplete prediction results. Summary of the invention

[0005] 1. Technical issues to be resolved In view of the shortcomings of the prior art, the present invention provides a traffic flow induction monitoring method and system based on convolutional denoising autoencoder, which collects and fuses multi-source heterogeneous traffic data, uses convolutional denoising autoencoder to denoise the spatiotemporal traffic data of vehicles with unknown navigation information, and then combines spatiotemporal convolutional neural network and long short-term memory network for feature extraction and sequence modeling, thereby obtaining a high-precision traffic flow prediction model. On this basis, by introducing a joint loss function, the loss of vehicles with known navigation information and vehicles with unknown navigation information is comprehensively considered, and the model is trained and optimized to improve the prediction performance. Finally, according to the prediction results and the traffic capacity of the road section, the driver is provided with the optimal driving path suggestion to achieve traffic flow induction and optimization.

[0006] (II) Technical solution To achieve the above objectives, the present invention is implemented by the following technical scheme: a traffic flow induction monitoring method based on convolutional denoising autoencoder, comprising the following steps: Collect video data, radar data, vehicle data, and navigation data of moving vehicles, match them according to timestamps and geographic locations, and generate a fused spatiotemporal traffic data set; Extract vehicle-mounted and navigation data from the spatiotemporal traffic dataset, and preliminarily divide vehicles into vehicles with known navigation information and vehicles with unknown navigation information. For vehicles with known navigation information, determine whether their current position deviates from the navigation path. If so, reclassify them as vehicles with unknown navigation information, otherwise do not process them; The spatiotemporal traffic data of vehicles with unknown navigation information are screened out from the spatiotemporal traffic data set, and the spatiotemporal traffic data of vehicles with unknown navigation information are processed using convolutional denoising autoencoders, spatiotemporal convolutional neural networks, and long short-term memory networks to generate a traffic flow prediction model; Extracting the spatiotemporal traffic data of vehicles with known navigation information from the spatiotemporal traffic data set, generating a joint loss function through the loss functions of vehicles with known navigation information and vehicles with unknown navigation information, and using the joint loss function and the spatiotemporal traffic data of vehicles with known navigation information to train and optimize the traffic flow prediction model; Based on the spatiotemporal traffic data of vehicles with known navigation information and vehicles with unknown navigation information, the traffic flow of the next road section is obtained. Combined with the saturation and traffic capacity of the next road section, the traffic index of the next road section is calculated. If the traffic index of the next road section is greater than the preset traffic threshold, the vehicle is guided to switch to the recommended alternative route.

[0007] Furthermore, the vehicles are initially divided into vehicles with known navigation information and vehicles with unknown navigation information, including: Extract the vehicle-mounted data and navigation data of all vehicles from the fused spatiotemporal traffic dataset. The vehicle-mounted data includes the vehicle's unique identifier and real-time location information, and the navigation data includes the vehicle's departure point, destination, and planned navigation path information; Based on whether the vehicle is associated with valid navigation data, all vehicles are initially divided into vehicles with known navigation information and vehicles with unknown navigation information. Vehicles with known navigation information are vehicles whose navigation data contains navigation path information, and vehicles with unknown navigation information are vehicles that are not associated with navigation data or whose navigation data does not contain navigation path information.

[0008] Further, determining whether the current position deviates from the navigation path includes: A path deviation threshold is set in advance. If the shortest distance between the vehicle's current position and the navigation path exceeds the path deviation threshold, the vehicle is judged to have deviated from the navigation path, and the vehicle is removed from the list of vehicles with known navigation information and reclassified as a vehicle with unknown navigation information; if the shortest distance between the vehicle's current position and the navigation path does not exceed the path deviation threshold, the vehicle is judged to have not deviated from the navigation path.

[0009] Furthermore, a traffic flow prediction model is generated, including: The spatiotemporal traffic data of vehicles with unknown navigation information are screened out from the spatiotemporal traffic data set, including the vehicle's video data, radar data, and vehicle-mounted data. The spatiotemporal traffic data of vehicles with unknown navigation information are denoised using a convolutional denoising autoencoder. The denoised spatiotemporal traffic data of vehicles with unknown navigation information are used as input, and high-order spatiotemporal features are extracted using a spatiotemporal convolutional neural network. The long short-term memory network is connected to perform sequence modeling on the extracted high-order spatiotemporal features; the output features of the spatiotemporal convolutional neural network and the long short-term memory network are spliced ​​to fuse the feature information from different network layers, and the fused features are sent to the fully connected layer for prediction to obtain a preliminary traffic flow prediction model.

[0010] Furthermore, the traffic flow prediction model is trained and optimized using a joint loss function, including: Calculate the proportion of vehicles with known navigation information among all vehicles on the road: p = number of vehicles with known navigation information / number of all vehicles on the road. Set a loss function for vehicles with unknown navigation information and vehicles with known navigation information respectively, and assign a weight coefficient to the loss function of vehicles with known navigation information: , where ω represents the weight coefficient and p represents the proportion of vehicles with known navigation information among all traveling vehicles.

[0011] Furthermore, the loss functions of vehicles with unknown navigation information and vehicles with known navigation information, as well as weight coefficients, are combined to generate a joint loss function: , where L_co represents the joint loss function, L_un represents the loss function of the vehicle with unknown navigation information, and L_kn represents the loss function of the vehicle with known navigation information; The traffic flow prediction model is trained using the joint loss function and the spatiotemporal traffic data of vehicles with known navigation information, and the traffic flow prediction model parameters are updated through the back propagation algorithm until the preset number of training rounds is reached or the loss function converges.

[0012] Furthermore, the traffic volume of the next road section is obtained, including: When the vehicle starts the on-board navigation or navigation software, the spatiotemporal traffic data of all vehicles with known navigation information are used to count the traffic flow of all vehicles with known navigation information when the current vehicle is proceeding to the next section. The spatiotemporal traffic data of vehicles with unknown navigation information is used as input, and the traffic flow prediction model is used to predict the traffic flow of vehicles with unknown navigation information on the next section.

[0013] Furthermore, the traffic index of the next road section is calculated, including: The traffic index of the next road section is calculated by combining the traffic flow of vehicles with known navigation information and vehicles with unknown navigation information with the saturation and traffic capacity of the next road section. The calculation formula is as follows: ; Among them, TI represents the traffic index, Tv_u represents the traffic flow of vehicles with known navigation information, Tv_k represents the traffic flow of vehicles with unknown navigation information, Se represents saturation, Tc represents traffic capacity, and Ct represents the congestion threshold.

[0014] Furthermore, when the traffic index of the next road segment is less than or equal to the preset traffic threshold, no processing is performed. When the traffic index of the next road segment is greater than the preset traffic threshold, alternative road segments are searched and the traffic indexes of all alternative road segments are calculated. If the traffic indexes of all alternative road segments are greater than the preset traffic threshold, the road segment with the lowest traffic index is selected as the recommended road segment, and a navigation instruction is sent to the vehicle to guide the vehicle to switch to the recommended alternative route.

[0015] Traffic flow induction monitoring system based on convolutional denoising autoencoder, including: The data acquisition module collects video data, radar data, vehicle data, and navigation data of moving vehicles, matches them according to timestamps and geographic locations, and generates a fused spatiotemporal traffic data set; The vehicle classification module extracts vehicle-mounted and navigation data from the spatiotemporal traffic data set, and preliminarily divides the moving vehicles into vehicles with known navigation information and vehicles with unknown navigation information. For vehicles with known navigation information, it determines whether their current position deviates from the navigation path. If it deviates from the navigation path, it is reclassified as a vehicle with unknown navigation information, otherwise no processing is done; The model building module selects the spatiotemporal traffic data of vehicles with unknown navigation information from the spatiotemporal traffic data set, processes the spatiotemporal traffic data of vehicles with unknown navigation information using convolutional denoising autoencoders, spatiotemporal convolutional neural networks, and long short-term memory networks to generate a traffic flow prediction model; The model training and optimization module extracts the spatiotemporal traffic data of vehicles with known navigation information from the spatiotemporal traffic data set, generates a joint loss function through the loss functions of vehicles with known navigation information and vehicles with unknown navigation information, and uses the joint loss function and the spatiotemporal traffic data of vehicles with known navigation information to train and optimize the traffic flow prediction model; The traffic flow induction decision module obtains the traffic volume of the next road section based on the spatiotemporal traffic data of vehicles with known navigation information and vehicles with unknown navigation information, and calculates the traffic index of the next road section based on the saturation and traffic capacity of the next road section. If the traffic index of the next road section is greater than the preset traffic threshold, the vehicle is guided to switch to the recommended alternative route.

[0016] (III) Beneficial effects The present invention provides a traffic flow induction monitoring method and system based on convolutional denoising autoencoding, which has the following beneficial effects: (1) By extracting and classifying vehicle information from the fused spatiotemporal traffic dataset and distinguishing between vehicles with known navigation information and vehicles with unknown navigation information, we can more accurately grasp the vehicle's driving status and path planning. For vehicles with known navigation information, we can promptly detect and handle vehicle deviations from the navigation path through path matching and deviation judgment, thereby ensuring the accuracy and integrity of traffic data.

[0017] (2) By using CDAE to denoise the spatiotemporal traffic data of vehicles with unknown navigation information, the noise and outliers in the data can be effectively removed, and it is more accurate and stable. The influence of vehicles with known navigation information is eliminated, and a large amount of calculation is reduced. The output features of STCN and LSTM are concatenated to fuse the feature information from different network layers, which helps the model to comprehensively utilize multiple information sources for prediction.

[0018] (3) By introducing the spatiotemporal traffic data of vehicles with known navigation information and combining it with the data of vehicles with unknown navigation information for joint training, the model can learn more diverse traffic flow characteristics, thereby improving its generalization ability in new scenarios or unknown conditions. Since the data of two types of vehicles are combined and reasonable weight coefficients are assigned according to their proportion in all traveling vehicles, the model can consider various factors more comprehensively when making predictions, thereby improving the accuracy of the predictions.

[0019] (4) By real-time monitoring and prediction of traffic flow on each road section, combined with the vehicle's navigation information, it can provide real-time traffic guidance during vehicle driving, help vehicles choose the optimal driving route, avoid congestion, and improve travel efficiency. By using traffic flow prediction models and traffic index calculations, it can intelligently judge the traffic conditions of each road section and provide vehicles with intelligent navigation suggestions based on the prediction results and real-time traffic data, thereby reducing the time and cost losses caused by traffic congestion. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flow chart of the traffic flow induction monitoring method based on convolutional denoising autoencoding of the present invention; Figure 2 It is a schematic diagram of the steps of the traffic flow induction monitoring method based on convolutional denoising autoencoding of the present invention; Figure 3 It is a schematic diagram of the structure of the traffic flow induction monitoring system based on convolutional denoising autoencoding of the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] See also Figure 1-Figure 2 The present invention provides a traffic flow induction monitoring method based on convolutional denoising autoencoding, comprising the following steps: Step 1: Collect video data, radar data, vehicle data, and navigation data of moving vehicles, match them according to timestamps and geographic locations, and generate a fused spatiotemporal traffic dataset; The step 1 includes the following contents: Step 101: deploy high-definition video cameras and microwave radar sensors at key traffic locations (such as intersections, highway entrances, urban main roads, etc.) to collect video data and radar data, use smart vehicle terminals to collect vehicle data, and use navigation software to collect navigation data; The camera is used to capture vehicle images, identify vehicle models, colors and license plate numbers. Considering the need for all-weather operation, the camera is equipped with infrared night vision function. The microwave radar sensor is used to measure dynamic traffic data such as vehicle speed, traffic volume and vehicle spacing. The radar uses frequency modulation continuous wave technology to improve detection accuracy. The sensor sampling frequency can reach the second level to meet the needs of real-time data collection. The intelligent vehicle terminal integrates GPS positioning and wireless communication modules to report the vehicle's GPS position, driving speed, etc. in real time. The navigation software is used to collect road condition information (such as congestion, road construction, etc.) as well as departure and destination information. Step 102: preprocessing the collected video data, radar data, vehicle data, and navigation data, including: using median filtering and weighted moving average methods to remove outliers and random noise, and using Z-score normalization method to unify data from different sensors to the same scale; Step 103: Use the 5G network to transmit the pre-processed data from each collection point to the data center, match the camera data, radar data, vehicle data, and navigation data according to timestamps and geographic locations, and generate a fused spatiotemporal traffic data set; When using, combine the contents of step 101 to step 103: By integrating video data, radar data, vehicle data and navigation data, we can obtain comprehensive information about traffic conditions, providing a rich data basis for traffic flow induction monitoring. The fusion of multiple data sources can complement each other's shortcomings and reduce the errors and uncertainties that may be caused by a single data source.

[0023] Step 2: Extract vehicle-mounted and navigation data from the spatiotemporal traffic dataset, and preliminarily divide the vehicles into vehicles with known navigation information and vehicles with unknown navigation information. For vehicles with known navigation information, determine whether their current position deviates from the navigation path. If they deviate from the navigation path, they are reclassified as vehicles with unknown navigation information, otherwise no processing is done; The step 2 includes the following contents: Step 201: extracting the vehicle-mounted data and navigation data of all vehicles from the fused spatiotemporal traffic data set. The vehicle-mounted data includes the vehicle's unique identifier (such as a license plate number or a device ID) and real-time location information (such as GPS coordinates) reported by the intelligent vehicle-mounted terminal. The navigation data includes the vehicle's departure point, destination, and planned navigation path information. Step 202: Based on whether the vehicle is associated with valid navigation data, all vehicles are preliminarily divided into vehicles with known navigation information and vehicles with unknown navigation information, wherein vehicles with known navigation information are vehicles whose navigation data contains navigation path information, and vehicles with unknown navigation information are vehicles that are not associated with navigation data or whose navigation data does not contain navigation path information; Step 203: For vehicles with known navigation information, path matching is performed using a path matching algorithm according to the real-time location information (GPS coordinates) and navigation path information reported by the vehicle. The path matching algorithm includes a Euclidean distance based on geographic coordinates, a Manhattan distance, or other matching algorithms; Step 204: pre-set a path deviation threshold to determine whether the current position has deviated from the navigation path. For each vehicle with known navigation information, if the shortest distance between the current position of the vehicle and the navigation path exceeds the path deviation threshold, the vehicle is determined to have deviated from the navigation path, and the vehicle is removed from the list of vehicles with known navigation information and reclassified as a vehicle with unknown navigation information. If the shortest distance between the current position of the vehicle and the navigation path does not exceed the path deviation threshold, no processing is performed. It should be noted that the path deviation threshold can be set by analyzing historical traffic data to determine the average deviation and maximum deviation between the vehicle and the navigation path during normal driving, so as to set a reasonable threshold range. The path deviation threshold should reflect as accurately as possible whether the vehicle has actually deviated from the navigation path; When used, combine the contents of step 201 to step 204: By extracting and classifying vehicle information from the fused spatiotemporal traffic dataset and distinguishing between vehicles with known navigation information and vehicles with unknown navigation information, we can more accurately grasp the vehicle's driving status and path planning. For vehicles with known navigation information, through path matching and deviation judgment, we can promptly discover and handle situations where the vehicle deviates from the navigation path, thereby ensuring the accuracy and completeness of traffic data.

[0024] Step 3: Filter out the spatiotemporal traffic data of vehicles with unknown navigation information from the spatiotemporal traffic dataset, use convolutional denoising autoencoder, spatiotemporal convolutional neural network and long short-term memory network to process the spatiotemporal traffic data of vehicles with unknown navigation information, and generate a traffic flow prediction model; The step three includes the following contents: Step 301: Filter out the spatiotemporal traffic data of vehicles with unknown navigation information from the spatiotemporal traffic data set, including the video data, radar data and vehicle-mounted data of the vehicle, and use a convolutional denoising autoencoder (CDAE) to denoise the spatiotemporal traffic data of vehicles with unknown navigation information; Among them, a CDAE model is constructed, including a convolutional encoder and a convolutional decoder. The encoder maps the input data to a low-dimensional latent space, and the decoder restores the latent representation to clean output data. Random noise (such as Gaussian noise, salt noise, etc.) is added to the input data, and the CDAE network is trained to minimize the reconstruction error between the noisy data and the original data. After the training is completed, the final CDAE model is obtained; Step 302: Taking the denoised spatiotemporal traffic data of the unknown navigation information vehicle as input, and using a spatial-temporal convolutional neural network (STCN) to extract high-order spatiotemporal features; STCN uses graph convolutional networks in the spatial dimension to capture the topological structure of the road network, uses causal convolution in the temporal dimension to achieve unidirectional temporal information transmission to avoid using future data, and uses Inception modules at the bottom of the network to increase the receptive field and capture more contextual information. Residual connections are used at the top layer to speed up model convergence and prevent overfitting. Step 303: After STCN, a long short-term memory network (LSTM) is connected to perform sequence modeling on the extracted high-order spatiotemporal features; It should be noted that LSTM can capture long-term dependencies by adaptively selecting information to be remembered and forgotten through a gating mechanism. On the basis of LSTM, an attention mechanism can also be introduced to dynamically adjust the weights of different time steps to highlight the impact of key moments. Step 304: concatenate the output features of STCN and LSTM to fuse the feature information from different network layers, and send the fused features to the fully connected layer for prediction to obtain a preliminary traffic flow prediction model to predict the traffic flow of each road section in the future time period (e.g., within 15-60 minutes), including vehicle flow, average speed and congestion index; It should be noted that the fully connected layer can adopt a recursive prediction structure, that is, first predict the traffic conditions (such as traffic volume, average speed, congestion index, etc.) of the first time step in the future, and then use it as input to predict the traffic conditions of the next time step, and so on, until the traffic conditions within the next 15-60 minutes are predicted; When used, combine the contents of step 301 to step 304: By using CDAE to denoise the spatiotemporal traffic data of vehicles with unknown navigation information, the noise and outliers in the data can be effectively removed, and it is more accurate and stable, eliminating the influence of vehicles with known navigation information, reducing a lot of calculations, and concatenating the output features of STCN and LSTM to fuse feature information from different network layers. This helps the model to make comprehensive use of multiple information sources for prediction. When predicting through the fully connected layer, key indicators such as vehicle flow, average speed and congestion index of each section in the future time period can be output, providing strong decision support for traffic management and induction.

[0025] Step 4: Extract the spatiotemporal traffic data of vehicles with known navigation information from the spatiotemporal traffic data set, generate a joint loss function through the loss functions of vehicles with known navigation information and vehicles with unknown navigation information, and use the joint loss function and the spatiotemporal traffic data of vehicles with known navigation information to train and optimize the traffic flow prediction model; The step 4 includes the following contents: Step 401: extracting the spatiotemporal traffic data of vehicles with known navigation information from the spatiotemporal traffic data set, including the video data, radar data and vehicle-mounted data of the vehicles, and dividing the spatiotemporal traffic data of vehicles with known navigation information into a training set, a validation set and a test set to ensure the randomness and representativeness of the data division; Step 402: Use the traffic flow prediction model trained in step 3 (based on the data of vehicles with unknown navigation information) to predict the test set of vehicles with known navigation information, and evaluate the performance indicators of the traffic flow prediction model according to the prediction results, including accuracy, recall rate, F1 score, mean square error (MSE), etc.; Step 403: Calculate the proportion of vehicles with known navigation information among all vehicles on the road: p = number of vehicles with known navigation information / number of all vehicles on the road. Set a loss function (such as mean square error MSE, cross entropy loss, etc.) for vehicles with unknown navigation information and vehicles with known navigation information respectively, and assign a weight coefficient to the loss function of vehicles with known navigation information: , where ω represents the weight coefficient, and p represents the proportion of vehicles with known navigation information among all vehicles on the road; Step 404: Combine the loss functions of the vehicle with unknown navigation information and the vehicle with known navigation information, as well as the weight coefficients, to generate a joint loss function: , where L_co represents the joint loss function, L_un represents the loss function of the vehicle with unknown navigation information, and L_kn represents the loss function of the vehicle with known navigation information; Step 405: Use the joint loss function to train the traffic flow prediction model, update the traffic flow prediction model parameters through the back propagation algorithm, until the preset number of training rounds is reached or the loss function converges, evaluate the performance of the optimized model on the validation set, and adjust the model parameters (such as learning rate, batch size, number of network layers, etc.) according to the evaluation results to obtain better model performance; When used, combine the contents of step 401 to step 405: By introducing the spatiotemporal traffic data of vehicles with known navigation information and combining it with the data of vehicles with unknown navigation information for joint training, the model can learn more diverse traffic flow characteristics, thereby improving its generalization ability in new scenarios or unknown conditions. Since the data of two types of vehicles are combined and reasonable weight coefficients are assigned according to their proportion in all traveling vehicles, the model can consider various factors more comprehensively when making predictions, thereby improving the accuracy of the predictions.

[0026] Step 5: Based on the spatiotemporal traffic data of vehicles with known navigation information and vehicles with unknown navigation information, the traffic flow of the next road section is obtained, and the traffic index of the next road section is calculated in combination with the saturation and traffic capacity of the next road section. If the traffic index of the next road section is greater than the preset traffic threshold, the vehicle is guided to switch to the recommended alternative route.

[0027] The step five includes the following contents: Step 501: when the vehicle starts the in-vehicle navigation or navigation software, the spatiotemporal traffic data of all vehicles with known navigation information are used to count the traffic volume of all vehicles with known navigation information when the current vehicle is on the next road section, and the data of vehicles with unknown navigation information is used as input to predict the traffic volume of vehicles with unknown navigation information on the next road section using a traffic flow prediction model; It should be noted that when the navigation information of the vehicle is known, the time it takes for the vehicle to arrive at the next road section can be calculated through the vehicle's current position, driving speed and the distance of each road section. Therefore, the time it takes for all vehicles with known navigation information to enter the next road section can be calculated, and the number of vehicles with known navigation information that enter the next road section at the same time as the current vehicle can be counted; Step 502: Calculate the traffic index of the next road segment by combining the traffic volume of vehicles with known navigation information and vehicles with unknown navigation information with the saturation and traffic capacity of the next road segment. The calculation formula is as follows: ; Among them, TI represents the traffic index, Tv_u represents the traffic flow of vehicles with known navigation information, Tv_k represents the traffic flow of vehicles with unknown navigation information, Se represents saturation, Tc represents traffic capacity, and Ct represents the congestion threshold; It should be noted that the traffic capacity refers to the maximum traffic volume that can pass through the next road section in a unit time, which is a fixed road parameter. The saturation refers to the maximum traffic volume that the next road section can bear under normal operation, but it usually does not exceed the traffic capacity. It reflects the degree of congestion on the road and is a value between the current traffic volume and the traffic capacity. The congestion threshold refers to the traffic volume value when the next road section begins to be obviously congested, which is an empirical value or a threshold determined based on historical data. Step 503: pre-set a traffic threshold. When the traffic index of the next road segment is less than or equal to the preset traffic threshold, no processing is performed. When the traffic index of the next road segment is greater than the preset traffic threshold, it is considered that the road segment may be congested or have poor traffic, and alternative road segments are searched, and the traffic indexes of all the alternative road segments are calculated. It should be noted that the setting of the traffic threshold needs to take into account the normal traffic capacity of the road. Under normal circumstances, the road should be able to accommodate a certain number of vehicles to pass smoothly. This amount is the traffic capacity of the road. The traffic threshold should be set within a certain proportion of the road capacity to ensure that the road can remain unobstructed in most cases. Step 504: If the traffic indexes of all candidate road sections are greater than the preset traffic threshold, the road section with the lowest traffic index is selected as the recommended road section, and a navigation instruction is sent to the vehicle to guide the vehicle to switch to the recommended candidate route; When used, combine the contents of step 501 to step 504: By real-time monitoring and prediction of traffic flow on each road section, combined with the vehicle's navigation information, it can provide real-time traffic guidance during vehicle driving, help vehicles choose the best driving route, avoid congestion, and improve travel efficiency. By using traffic flow prediction models and traffic index calculations, it can intelligently judge the traffic conditions of each road section, and provide vehicles with intelligent navigation suggestions based on the prediction results and real-time traffic data, reducing time and cost losses caused by traffic congestion.

[0028] See also Figure 3 The present invention also provides a traffic flow induction monitoring system based on convolutional denoising autoencoder, including: a data acquisition module, a vehicle classification module, a model building module, a model training and optimization module and a traffic flow induction decision module; wherein, The data acquisition module collects video data, radar data, vehicle data, and navigation data of moving vehicles, matches them according to timestamps and geographic locations, and generates a fused spatiotemporal traffic data set; The vehicle classification module extracts vehicle-mounted and navigation data from the spatiotemporal traffic data set, and preliminarily divides the moving vehicles into vehicles with known navigation information and vehicles with unknown navigation information. For vehicles with known navigation information, it determines whether their current position deviates from the navigation path. If it deviates from the navigation path, it is reclassified as a vehicle with unknown navigation information, otherwise no processing is done; The model building module selects the spatiotemporal traffic data of vehicles with unknown navigation information from the spatiotemporal traffic data set, processes the spatiotemporal traffic data of vehicles with unknown navigation information using convolutional denoising autoencoders, spatiotemporal convolutional neural networks, and long short-term memory networks to generate a traffic flow prediction model; The model training and optimization module extracts the spatiotemporal traffic data of vehicles with known navigation information from the spatiotemporal traffic data set, generates a joint loss function through the loss functions of vehicles with known navigation information and vehicles with unknown navigation information, and uses the joint loss function and the spatiotemporal traffic data of vehicles with known navigation information to train and optimize the traffic flow prediction model; The traffic flow induction decision module obtains the traffic volume of the next road section based on the spatiotemporal traffic data of vehicles with known navigation information and vehicles with unknown navigation information, and calculates the traffic index of the next road section based on the saturation and traffic capacity of the next road section. If the traffic index of the next road section is greater than the preset traffic threshold, the vehicle is guided to switch to the recommended alternative route.

[0029] In the application, the several formulas involved are all calculated by removing dimensions and taking their numerical values. The formula is a formula of the most recent real situation obtained by collecting a large amount of data and performing software simulation. The coefficients in the formula are set by technicians in this field according to actual conditions.

[0030] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented in combination with electronic hardware, computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0031] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0032] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A traffic flow induction monitoring method based on convolutional denoising autoencoder, characterized by: The following steps are involved: Collect video data, radar data, vehicle data, and navigation data of moving vehicles, match them according to timestamps and geographic locations, and generate a fused spatiotemporal traffic data set; Extract vehicle-mounted and navigation data from the spatiotemporal traffic dataset, and preliminarily divide the vehicles into vehicles with known navigation information and vehicles with unknown navigation information. For vehicles with known navigation information, determine whether their current position deviates from the navigation path. If they deviate from the navigation path, they are reclassified as vehicles with unknown navigation information, otherwise no processing is done; The spatiotemporal traffic data of vehicles with unknown navigation information are screened out from the spatiotemporal traffic data set, and the spatiotemporal traffic data of vehicles with unknown navigation information are processed using convolutional denoising autoencoders, spatiotemporal convolutional neural networks, and long short-term memory networks to generate a traffic flow prediction model; Extracting the spatiotemporal traffic data of vehicles with known navigation information from the spatiotemporal traffic data set, generating a joint loss function through the loss functions of vehicles with known navigation information and vehicles with unknown navigation information, and using the joint loss function and the spatiotemporal traffic data of vehicles with known navigation information to train and optimize the traffic flow prediction model; Based on the spatiotemporal traffic data of vehicles with known navigation information and vehicles with unknown navigation information, the traffic flow of the next road section is obtained. Combined with the saturation and traffic capacity of the next road section, the traffic index of the next road section is calculated. If the traffic index of the next road section is greater than the preset traffic threshold, the vehicle is guided to switch to the recommended alternative route.

2. The traffic flow induction monitoring method based on convolutional denoising autoencoder according to claim 1 is characterized in that: The vehicles are initially divided into vehicles with known navigation information and vehicles with unknown navigation information, including: Extract the vehicle-mounted data and navigation data of all vehicles from the fused spatiotemporal traffic dataset. The vehicle-mounted data includes the vehicle's unique identifier and real-time location information, and the navigation data includes the vehicle's departure point, destination, and planned navigation path information; Based on whether the vehicle is associated with valid navigation data, all vehicles are initially divided into vehicles with known navigation information and vehicles with unknown navigation information. Vehicles with known navigation information are vehicles whose navigation data contains navigation path information, and vehicles with unknown navigation information are vehicles that are not associated with navigation data or whose navigation data does not contain navigation path information.

3. The traffic flow induction monitoring method based on convolutional denoising autoencoder according to claim 2 is characterized in that: For a vehicle with known navigation information, determine whether its current position deviates from the navigation path, including: A path deviation threshold is set in advance. If the shortest distance between the current position of a vehicle with known navigation information and the navigation path exceeds the path deviation threshold, the vehicle is judged to have deviated from the navigation path, and the vehicle is removed from the list of vehicles with known navigation information and reclassified as a vehicle with unknown navigation information; if the shortest distance between the current position of the vehicle and the navigation path does not exceed the path deviation threshold, the vehicle is judged to have not deviated from the navigation path.

4. The traffic flow induction monitoring method based on convolutional denoising autoencoder according to claim 1 is characterized in that: Generate traffic flow prediction models, including: The spatiotemporal traffic data of vehicles with unknown navigation information are screened out from the spatiotemporal traffic data set, including the vehicle's video data, radar data, and vehicle-mounted data. The spatiotemporal traffic data of vehicles with unknown navigation information are denoised using a convolutional denoising autoencoder. The denoised spatiotemporal traffic data of vehicles with unknown navigation information are used as input, and high-order spatiotemporal features are extracted using a spatiotemporal convolutional neural network. The long short-term memory network is connected to perform sequence modeling on the extracted high-order spatiotemporal features; the output features of the spatiotemporal convolutional neural network and the long short-term memory network are spliced ​​to fuse the feature information from different network layers, and the fused features are sent to the fully connected layer for prediction to obtain a preliminary traffic flow prediction model.

5. The traffic flow induction monitoring method based on convolutional denoising autoencoder according to claim 1 is characterized in that: The traffic flow prediction model is trained and optimized using a joint loss function, including: Calculate the proportion of vehicles with known navigation information among all vehicles on the road: p = number of vehicles with known navigation information / number of all vehicles on the road. Set a loss function for vehicles with unknown navigation information and vehicles with known navigation information respectively, and assign a weight coefficient to the loss function of vehicles with known navigation information: , where ω represents the weight coefficient and p represents the proportion of vehicles with known navigation information among all traveling vehicles.

6. The traffic flow induction monitoring method based on convolutional denoising autoencoder according to claim 5 is characterized in that: The joint loss function is generated by the loss functions of vehicles with unknown navigation information and vehicles with known navigation information, as well as the weight coefficients: , where L_co represents the joint loss function, L_un represents the loss function of the vehicle with unknown navigation information, and L_kn represents the loss function of the vehicle with known navigation information; The traffic flow prediction model is trained using the joint loss function and the spatiotemporal traffic data of vehicles with known navigation information, and the traffic flow prediction model parameters are updated through the back propagation algorithm until the preset number of training rounds is reached or the loss function converges.

7. The traffic flow induction monitoring method based on convolutional denoising autoencoder according to claim 1 is characterized in that: Get the traffic volume of the next road segment, including: When the vehicle starts the on-board navigation or navigation software, the spatiotemporal traffic data of all vehicles with known navigation information are used to count the traffic flow of all vehicles with known navigation information when the current vehicle is proceeding to the next section. The spatiotemporal traffic data of vehicles with unknown navigation information is used as input, and the traffic flow prediction model is used to predict the traffic flow of vehicles with unknown navigation information on the next section.

8. The traffic flow induction monitoring method based on convolutional denoising autoencoder according to claim 7 is characterized in that: Calculate the traffic index of the next road segment, including: The traffic index of the next road section is calculated by combining the traffic flow of vehicles with known navigation information and vehicles with unknown navigation information with the saturation and traffic capacity of the next road section. The calculation formula is as follows: ； Among them, TI represents the traffic index, Tv_u represents the traffic flow of vehicles with known navigation information, Tv_k represents the traffic flow of vehicles with unknown navigation information, Se represents saturation, Tc represents traffic capacity, and Ct represents the congestion threshold.

9. The traffic flow induction monitoring method based on convolutional denoising autoencoder according to claim 8 is characterized in that: When the traffic index of the next road segment is less than or equal to the preset traffic threshold, no processing is performed. When the traffic index of the next road segment is greater than the preset traffic threshold, alternative road segments are searched and the traffic indexes of all alternative road segments are calculated. If the traffic indexes of all candidate sections are greater than the preset traffic threshold, the section with the lowest traffic index is selected as the recommended section, and a navigation instruction is sent to the vehicle to guide the vehicle to switch to the recommended alternative route.

10. A traffic flow induction monitoring system based on convolutional denoising autoencoder, used to implement the method according to any one of claims 1 to 9, characterized in that: include: The data acquisition module collects video data, radar data, vehicle data, and navigation data of moving vehicles, matches them according to timestamps and geographic locations, and generates a fused spatiotemporal traffic data set; The vehicle classification module extracts vehicle-mounted and navigation data from the spatiotemporal traffic data set, and preliminarily divides the moving vehicles into vehicles with known navigation information and vehicles with unknown navigation information. For vehicles with known navigation information, it determines whether their current position deviates from the navigation path. If it deviates from the navigation path, it is reclassified as a vehicle with unknown navigation information, otherwise no processing is done; The model building module selects the spatiotemporal traffic data of vehicles with unknown navigation information from the spatiotemporal traffic data set, processes the spatiotemporal traffic data of vehicles with unknown navigation information using convolutional denoising autoencoders, spatiotemporal convolutional neural networks, and long short-term memory networks to generate a traffic flow prediction model; The model training and optimization module extracts the spatiotemporal traffic data of vehicles with known navigation information from the spatiotemporal traffic data set, generates a joint loss function through the loss functions of vehicles with known navigation information and vehicles with unknown navigation information, and uses the joint loss function and the spatiotemporal traffic data of vehicles with known navigation information to train and optimize the traffic flow prediction model; The traffic flow induction decision module obtains the traffic volume of the next road section based on the spatiotemporal traffic data of vehicles with known navigation information and vehicles with unknown navigation information, and calculates the traffic index of the next road section based on the saturation and traffic capacity of the next road section. If the traffic index of the next road section is greater than the preset traffic threshold, the vehicle is guided to switch to the recommended alternative route.

Images