Traffic control and guidance system and method based on ant colony algorithm

Abstract

This invention discloses a traffic control and guidance system and method based on ant colony optimization, belonging to the field of road traffic control technology. It addresses the problem that existing methods, particularly neural network models, primarily focus on the individual route conditions of multiple branching routes at intersections on one-way roads, resulting in low information fusion between various traffic elements. The method includes identifying dynamic vehicle characteristics within the road network, pre-constructing a road network optimization model based on ant colony optimization combined with deep learning, and performing cluster solution decision analysis on the pheromone diffusion function. In this invention, by pre-constructing a road network optimization model based on ant colony optimization combined with deep learning, the coordinated optimization of traffic signal control and vehicle guidance is achieved. By outputting optimized traffic signal control strategies and optimized vehicle path strategies, signal timing and guidance paths can work together to improve the operational efficiency of the traffic system and ensure the accuracy and effectiveness of the guidance strategy.

Description

Technical Field

[0001] This invention belongs to the field of road traffic control technology, specifically relating to a traffic control and guidance system and method based on ant colony algorithm. Background Technology

[0002] With rapid economic development and accelerated urbanization, the number of motor vehicles in cities has increased dramatically. Traffic congestion has become a common problem in major cities worldwide, not only significantly increasing residents' travel time and reducing their quality of life, but also causing huge economic losses, such as fuel waste and accelerated vehicle wear and tear. During peak commuting hours, major roads, intersections, and other traffic hubs are often completely congested, with vehicles queuing for excessively long periods at red lights, and the instability of traffic flow can also lead to a series of traffic accident hazards.

[0003] Traffic guidance systems aim to provide drivers with route planning suggestions, guiding vehicles to be rationally distributed across different roads. However, many current guidance systems rely on insufficiently timely updates to traffic information, and their route planning algorithms are relatively simple, often based solely on shortest paths or fixed weights, failing to fully consider complex factors such as the dynamic evolution of traffic flow and the propagation characteristics of road congestion. When faced with sudden traffic accidents or large-scale events causing a surge in localized traffic volume, these systems cannot quickly and accurately replan guidance routes, resulting in some roads becoming oversaturated while other road resources remain idle and wasted.

[0004] Chinese patent CN114999191B discloses an urban traffic guidance method and system. The method includes: acquiring the route information of multiple branching routes at a fork in a one-way road within the city; generating guidance reminders for each of the multiple branching routes based on a pre-set neural network model; and providing guidance reminders to traffic flow about to pass through the fork based on the guidance reminders for each of the multiple branching routes. However, the existing method's neural network model mainly focuses on the route information of multiple branching routes at a fork in a one-way road, which has a relatively limited scope and cannot simulate dynamic vehicle behavior. The generation of its guidance reminders may rely more on static route analysis, and the information fusion between various traffic elements is low. To address these issues, we propose a traffic control and guidance system and method based on the ant colony algorithm. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a traffic control and guidance system and method based on ant colony algorithm. This solves the problems of existing methods where neural network models mainly focus on the individual route conditions of multiple branching routes at a fork in a one-way road, resulting in a relatively limited scope of consideration. These methods cannot simulate dynamic vehicle behavior, and the generation of guidance and reminder content may rely more on static route condition analysis, leading to low information fusion between various traffic elements.

[0006] This invention is implemented as follows: a traffic control and guidance method based on ant colony algorithm, wherein the traffic control and guidance method based on ant colony algorithm includes:

[0007] Real-time traffic flow data of road network is acquired based on computer vision, and the real-time traffic flow data of road network is analyzed to identify the dynamic characteristics of vehicles in the road network.

[0008] A road network optimization model based on ant colony algorithm combined with deep learning is pre-constructed. Vehicle dynamic feature information is used as the initial pheromone to give the road network optimization model. The dynamic behavior of vehicles is simulated based on the cell transmission algorithm. The dynamic behavior of vehicles is used as the heuristic information of the main road segments of the road network. The pheromone diffusion model based on ant colony algorithm is constructed by combining the initial pheromone and the heuristic information of the main road segments of the road network.

[0009] Load the pheromone diffusion model, and with the pheromone update, vehicle path duration, and vehicle obstacle avoidance count as constraints, solve the pheromone diffusion function of the pheromone diffusion model to obtain the cluster solution of the pheromone diffusion function.

[0010] Obtain cluster solutions of the pheromone diffusion function, and combine dynamic network adjustment algorithm and multi-objective optimization algorithm to conduct decision analysis on the cluster solutions of the pheromone diffusion function, and output traffic signal optimization control strategy and vehicle optimization path strategy.

[0011] Preferably, the method for identifying vehicle dynamic characteristic information within the road network includes:

[0012] Load real-time traffic flow data of the road network, clean the real-time traffic flow data of the road network and handle outliers to obtain pre-processed real-time traffic flow data of the road network. The real-time traffic flow data of the road network includes vehicle identification, traffic volume of main / secondary road segments, vehicle speed, traffic light duration, vehicle initial / end position, and road segments passed through.

[0013] Based on the principal component analysis method, real-time traffic flow data of different types of road networks are weighted to obtain real-time weights of data types that represent the importance of data. The real-time weights of data types are then mapped to a pre-constructed data level evaluation mapping table, and at least one set of real-time vectors of data types are output.

[0014] Obtain at least one set of real-time vectors of different data types, and concatenate and fuse the at least one set of real-time vectors of different data types to obtain a real-time comprehensive vector representing vehicle dynamic information.

[0015] Identify the traffic volume, speed, traffic light duration, and route segments of the main / secondary road segments in the real-time traffic flow data of the road network. Using the complex network method, road segment intersections are abstracted as nodes of the traffic network, and vehicles on the main / secondary road segments are abstracted as connecting edges of the traffic network nodes, thus obtaining a dynamic directed traffic network.

[0016] Traffic flow, speed, and traffic light duration data of the main / secondary road segments in the dynamic directed traffic network are extracted. Based on the seepage analysis mechanism, the traffic flow, speed, and traffic light duration data of the main / secondary road segments are normalized to obtain traffic flow matrix, speed matrix, and traffic light duration matrix. The traffic flow matrix, speed matrix, and traffic light duration matrix are superimposed to obtain the multi-scale fusion matrix of the main / secondary road segments.

[0017] The DDPG algorithm, a deep learning algorithm, is used to calculate the multi-scale fusion matrix of the main / secondary road sections to obtain the transient seepage index that characterizes the transient congestion level of the main / secondary road sections.

[0018] By associating the road segments in the real-time traffic flow data with transient seepage indices, the real-time comprehensive vector is corrected to obtain the corrected comprehensive vector. The corrected comprehensive vector is then used as the dynamic characteristic information of vehicles within the road network and output.

[0019] Preferably, the method for pre-constructing a road network optimization model based on ant colony optimization combined with deep learning includes:

[0020] A hybrid infrastructure for the road network optimization model is adopted, using the spatiotemporal prediction model ST-GCN and the sequence prediction model LSTM. An autoencoder is introduced before the hybrid infrastructure, and a feature fusion module based on deep learning is introduced after the hybrid infrastructure. The feature fusion module has a built-in ant colony optimization engine, which incorporates the ant colony algorithm. A multi-scale optimization module is introduced after the feature fusion module. The multi-scale optimization module includes a signal optimization layer and a path guidance layer. A classification decision layer is set after the multi-scale optimization module. The classification decision layer is used to output traffic signal optimization control strategy and vehicle path optimization strategy.

[0021] Capture historical traffic flow data of the road network, analyze the historical traffic flow data of the road network, identify historical dynamic feature information within the road network, and divide the historical dynamic feature information into training set and test set;

[0022] Define the loss functions and optimizers for the autoencoder, hybrid infrastructure, feature fusion module, and multi-scale optimization module respectively, and preset the training epochs and hyperparameters;

[0023] The training set is used to pre-train the autoencoder based on unsupervised learning, extracting the spatiotemporal sequence features after encoding. Gaussian noise is added during training to enhance the robustness of the autoencoder. After pre-training, the parameters of the autoencoder are frozen. The spatiotemporal sequence features extracted by the autoencoder are fused with the training set and input into the hybrid infrastructure. The loss between the predicted output and the true value is calculated. Backpropagation is used to update the weights of the spatiotemporal prediction model and the sequence prediction model. The iteration is repeated until the loss of the hybrid infrastructure converges.

[0024] The spatiotemporal sequence features output by the hybrid infrastructure are loaded and used as initial pheromones. A self-supervised reinforcement learning mechanism is introduced to jointly train the feature fusion module and the multi-scale optimization module. During the joint training of the feature fusion module and the multi-scale optimization module, the weights and parameters of the feature fusion module and the multi-scale optimization module are dynamically adjusted through the ant colony optimization engine. The converged feature fusion module and multi-scale optimization module are output. After the road network optimization model is trained, the difference between the classification result and the true result is determined by the multi-task joint loss function. The parameters of the road network optimization model are adjusted based on the end-to-end fine-tuning strategy. The iteration is repeated until the loss converges and the converged road network optimization model is output.

[0025] Obtain a test set, use the test set as input, execute the pre-trained road network optimization model, obtain the test results, calculate the prediction accuracy of the test results, determine whether the prediction accuracy meets the preset test accuracy threshold, and if the prediction accuracy meets the preset test accuracy threshold, output the converged road network optimization model.

[0026] Preferably, the method for simulating vehicle dynamic behavior based on cell transport algorithm includes:

[0027] The system acquires vehicle dynamic feature information, processes the vehicle dynamic feature information using an autoencoder, and outputs the encoded spatiotemporal sequence features.

[0028] The hybrid infrastructure performs spatiotemporal prediction and sequence prediction processing on spatiotemporal sequence features to obtain spatiotemporal sequence features containing spatiotemporal prediction results and sequence prediction results. The spatiotemporal sequence features are used as initial pheromones, and the feature fusion module dynamically corrects the initial pheromones based on the ant colony optimization engine to obtain pheromone corrections.

[0029] The dynamic directed traffic network is obtained, and information modifiers are used as heuristic information for the dynamic directed traffic network. The dynamic directed traffic network is initialized, and the road segments in the dynamic directed traffic network are divided into cell units of the same length. The maximum flow, upstream flow, downstream flow, and transmission rate of the cell units are updated based on the heuristic information of the dynamic directed traffic network.

[0030] The updated cell unit is loaded, and the cell unit is simulated based on the preset cell unit time step, vehicle transmission calculation parameters, traffic density threshold, and number of simulations to obtain vehicle dynamic behavior. The vehicle dynamic behavior is used as heuristic information for the main road segments of the road network. During the cell unit simulation, a pheromone diffusion model based on the ant colony algorithm is constructed by combining the initial pheromone and the heuristic information of the main road segments of the road network. The pheromone diffusion model is then optimized and the optimized pheromone diffusion model is output.

[0031] Load the pheromone diffusion model, and with constraints such as updating pheromones, vehicle path duration, and vehicle obstacle avoidance times, solve the pheromone diffusion function of the pheromone diffusion model to obtain the cluster solution of the pheromone diffusion function.

[0032] Preferably, the method for decision analysis of pheromone diffusion function cluster solutions combining dynamic network adjustment algorithm and multi-objective optimization algorithm includes:

[0033] The pheromone diffusion function of the pheromone diffusion model is loaded. With the goal of balancing traffic flow in the road network and optimizing the phase time and green light time of traffic signals, the pheromone diffusion function cluster solution is obtained by dynamically balancing the traffic flow in the road network, optimizing the phase time and green light time of traffic signals based on random bubble sort.

[0034] The parameters of the pheromone diffusion model are updated based on the dynamic network adjustment algorithm, and the fitness function is used to calculate the individual optimal position and the global optimal position of the vehicle during the iterative update based on the ant colony algorithm.

[0035] By combining a dynamic pheromone mapping mechanism to map vehicle dynamic behavior into ant colony behavior parameters, and using a multi-objective optimization algorithm to dynamically fuse the cluster solution of the pheromone diffusion function and the ant colony behavior parameters, at least one set of dynamic fused solutions is obtained. The dynamic fused solutions are sorted and filtered by the TOPSIS method to determine the final road network optimization strategy. The road network optimization strategy includes a traffic signal optimization control strategy and a vehicle optimization path strategy, and outputs the traffic signal optimization control strategy and the vehicle optimization path strategy.

[0036] On the other hand, the present invention also provides a traffic control and guidance system based on ant colony algorithm, the traffic control and guidance system based on ant colony algorithm comprising:

[0037] The traffic flow acquisition module acquires real-time traffic flow data of the road network based on computer vision, analyzes the real-time traffic flow data of the road network, and identifies the dynamic characteristics of vehicles within the road network.

[0038] The vehicle simulation module pre-constructs a road network optimization model based on ant colony algorithm combined with deep learning. Vehicle dynamic feature information is used as initial pheromone to give the road network optimization model. The module simulates vehicle dynamic behavior based on cell transmission algorithm and uses vehicle dynamic behavior as heuristic information for the main road segments of the road network. The module combines the initial pheromone and the heuristic information of the main road segments of the road network to construct a pheromone diffusion model based on ant colony algorithm.

[0039] The pheromone diffusion module is used to load the pheromone diffusion model, and with the constraints of updating pheromones, vehicle path duration, and vehicle obstacle avoidance times, solve the pheromone diffusion function of the pheromone diffusion model to obtain the cluster solution of the pheromone diffusion function.

[0040] The road network optimization module is used to obtain cluster solutions of the pheromone diffusion function, and to conduct decision analysis on the cluster solutions of the pheromone diffusion function by combining dynamic network adjustment algorithm and multi-objective optimization algorithm, and output traffic signal optimization control strategy and vehicle optimization path strategy.

[0041] Preferably, the traffic flow acquisition module includes:

[0042] The preprocessing unit is used to load real-time traffic flow data of the road network, clean the real-time traffic flow data of the road network and handle outliers to obtain preprocessed real-time traffic flow data of the road network.

[0043] The weight mapping unit assigns weights to real-time traffic flow data of different types of road networks based on principal component analysis, obtains real-time weights of data types that characterize the importance of the data, maps the real-time weights of data types to a pre-built data level evaluation mapping table, and outputs at least one set of real-time vectors of data types.

[0044] The traffic network construction unit is used to acquire at least one set of real-time vectors of data types, splice and fuse at least one set of real-time vectors of data types to obtain a real-time comprehensive vector representing vehicle dynamic information, identify the traffic flow, vehicle speed, traffic light duration and route segments of the main / secondary road segments in the real-time traffic flow data of the road network, and abstract the road segment intersections as nodes of the traffic network according to the complex network method, and abstract the main / secondary road segment vehicles as the connecting edges of the traffic network nodes to obtain a dynamic directed traffic network.

[0045] The multi-scale fusion unit is used to extract traffic flow, speed data and traffic light duration of the main / secondary road segments in the dynamic directed traffic network. Based on the seepage analysis mechanism, the traffic flow, speed data and traffic light duration of the main / secondary road segments are normalized to obtain the traffic flow matrix, speed matrix and light duration matrix. The traffic flow matrix, speed matrix and light duration matrix are superimposed to obtain the multi-scale fusion matrix of the main / secondary road segments.

[0046] The feature information output unit uses the deep learning DDPG algorithm to calculate the multi-scale fusion matrix of the main / secondary road segments, and obtains the transient seepage index that characterizes the transient congestion level of the main / secondary road segments. The transient seepage index is used to correct the real-time comprehensive vector based on the road segments in the real-time traffic flow data, and the corrected comprehensive vector is obtained. The corrected comprehensive vector is used as the dynamic feature information of vehicles in the road network and output.

[0047] Compared with the prior art, the embodiments of this application have the following main advantages:

[0048] In this embodiment of the invention, a pre-constructed road network optimization model based on ant colony optimization and deep learning is used to comprehensively perceive and analyze the traffic conditions of the entire road network. The full coverage of the road network allows the optimization model to comprehensively consider the mutual influence between various traffic nodes. Furthermore, the introduction of Cellular Transmission (CTM) algorithm to simulate the dynamic behavior of vehicles in the road network enables the optimization model to adjust the guidance strategy according to real-time traffic conditions, better cope with traffic flow fluctuations and emergencies, and combine dynamic network adjustment algorithm and multi-objective optimization algorithm to perform cluster solution decision analysis of pheromone diffusion function. Based on traffic flow data, the pheromone diffusion function and guidance strategy are dynamically adjusted, realizing the coordinated optimization of traffic signal control and vehicle guidance. By outputting traffic signal optimization control strategy and vehicle optimization path strategy, signal timing and guidance path can cooperate with each other to improve the operating efficiency of the traffic system and ensure the accuracy and effectiveness of the guidance strategy.

[0049] This invention provides a method for identifying dynamic vehicle feature information within a road network. During vehicle dynamic feature information identification, multi-dimensional data such as vehicle identification, traffic flow, vehicle speed, and traffic light duration are fused into a single-dimensional comprehensive vector. Timestamp alignment and temporal encoding are preserved, reducing computational complexity while maintaining spatiotemporal correlation. A transient seepage index is calculated in real-time based on a deep reinforcement learning algorithm to accurately quantify road segment congestion. Weighted correction of the comprehensive vector using the transient seepage index significantly reduces noise interference and improves feature robustness, thereby enhancing the accuracy and timeliness of vehicle dynamic feature information, making it more realistic, and providing a reliable basis for traffic control and guidance strategy formulation.

[0050] This invention provides a road network optimization model and its construction method. By constructing a hybrid infrastructure including a spatiotemporal prediction model ST-GCN and a sequence prediction model LSTM, and introducing autoencoders and feature fusion modules before and after, the model can fully mine and integrate the spatiotemporal and sequence features in road network traffic flow data. The effective feature extraction and dimensionality reduction of the autoencoder, along with the deep fusion processing of the feature fusion module, enables the model to more accurately capture the complex patterns and trends of traffic flow, thereby significantly improving the prediction accuracy of key traffic parameters such as traffic flow and vehicle speed. Furthermore, an ant colony optimization engine is introduced, combining ant colony algorithms with deep learning. Utilizing the global search capability and positive feedback mechanism of ant colony algorithms, the optimal traffic signal control strategy and vehicle path planning scheme are efficiently searched in complex road network environments. The road network optimization model can organically combine traffic signal optimization control with vehicle path guidance, achieving coordinated operation between the two. By optimizing traffic signal timing and vehicle travel paths, the utilization efficiency of road resources is improved, vehicle queuing time and delays are reduced, and energy consumption and environmental pollution are decreased.

[0051] In this embodiment of the invention, when simulating vehicle dynamic behavior based on the cell transmission algorithm, the high-dimensional vehicle dynamic feature data is reduced in dimensionality through the encoding process, making the data easier to process and analyze, while retaining the core information in the data. This helps to improve the training speed and prediction accuracy of the subsequent model. The feature fusion module dynamically corrects the initial pheromones based on the ant colony optimization engine. Combined with the global search capability and positive feedback mechanism of the ant colony algorithm, it can dynamically adjust the distribution and intensity of pheromones according to real-time traffic conditions, enabling the model to better adapt to changes in traffic flow, further improving the accuracy of prediction and optimization effect, and providing more realistic guidance information for vehicle dynamic behavior simulation. Attached Figure Description

[0052] Figure 1 This is a schematic diagram illustrating the implementation process of the traffic control and guidance method based on the ant colony algorithm provided by this invention.

[0053] Figure 2 A schematic diagram illustrating the implementation process of a method for identifying dynamic characteristics of vehicles within a road network is shown.

[0054] Figure 3 A schematic diagram illustrating the implementation process of a method for pre-constructing a road network optimization model based on ant colony algorithm combined with deep learning is shown.

[0055] Figure 4 A schematic diagram illustrating the implementation process of a method for simulating vehicle dynamics based on a cell transport algorithm is shown.

[0056] Figure 5A schematic diagram illustrates the implementation process of a method combining dynamic network adjustment algorithm and multi-objective optimization algorithm for cluster solution decision analysis of pheromone diffusion function.

[0057] Figure 6 A schematic diagram of a traffic control and guidance system based on the ant colony algorithm is shown. Detailed Implementation

[0058] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0059] Existing neural network models primarily focus on the individual route conditions of multiple branching paths at a fork in a one-way road. Their scope is relatively limited, failing to simulate dynamic vehicle behavior. Furthermore, the generation of guidance and reminder content may rely heavily on static route analysis, resulting in low information fusion between various traffic elements. To address these issues, we propose a traffic control and guidance system and method based on ant colony optimization (ACO). In short, the method first acquires real-time traffic flow data of the road network using computer vision, then analyzes this data to identify dynamic vehicle characteristics within the road network, and pre-constructs a road network based on ACO combined with deep learning. The optimization model incorporates vehicle dynamic characteristic information as initial pheromone into the road network optimization model. Based on the cell transmission algorithm, it simulates vehicle dynamic behavior and uses this behavior as heuristic information for the main road segments. Combining the initial pheromone and the heuristic information from the main road segments, it constructs a pheromone diffusion model based on the ant colony algorithm. Then, with constraints such as pheromone updates, vehicle path duration, and vehicle obstacle avoidance attempts, it solves the pheromone diffusion function of the model, obtaining a cluster solution for the pheromone diffusion function. Finally, it combines dynamic network adjustment algorithms and multi-objective optimization algorithms to conduct decision analysis on the cluster solution of the pheromone diffusion function, outputting traffic signal optimization control strategies and vehicle optimization path strategies. In this embodiment of the invention, a pre-constructed road network optimization model based on ant colony optimization and deep learning is used to comprehensively perceive and analyze the traffic conditions of the entire road network. The full coverage of the road network allows the optimization model to comprehensively consider the mutual influence between various traffic nodes. Furthermore, the introduction of Cellular Transmission (CTM) algorithm to simulate the dynamic behavior of vehicles in the road network enables the optimization model to adjust the guidance strategy according to real-time traffic conditions, better cope with traffic flow fluctuations and emergencies, and combine dynamic network adjustment algorithm and multi-objective optimization algorithm to perform cluster solution decision analysis of pheromone diffusion function. Based on traffic flow data, the pheromone diffusion function and guidance strategy are dynamically adjusted, realizing the coordinated optimization of traffic signal control and vehicle guidance. By outputting traffic signal optimization control strategy and vehicle optimization path strategy, signal timing and guidance path can cooperate with each other to improve the operating efficiency of the traffic system and ensure the accuracy and effectiveness of the guidance strategy.

[0060] This invention provides a traffic control and guidance method based on ant colony algorithm. Figure 1 A schematic diagram illustrating the implementation process of a traffic control and guidance method based on the ant colony algorithm is shown. The traffic control and guidance method based on the ant colony algorithm specifically includes:

[0061] Step S10: Obtain real-time traffic flow data of the road network based on computer vision, analyze the real-time traffic flow data of the road network, and identify the dynamic characteristic information of vehicles in the road network.

[0062] It's important to note that computer vision technology can capture real-time dynamic information such as the position, speed, and direction of vehicles in a road network. Compared to traditional fixed sensors (such as coils and radar), it offers wider coverage and richer data dimensions, making it particularly suitable for complex intersections and multi-lane scenarios. When acquiring real-time traffic flow data, computer vision utilizes image sensors, LiDAR, infrared thermal imagers, and webcams to collect multimodal real-time traffic data. Analyzing this data not only provides macroscopic traffic flow data but also allows for in-depth identification of dynamic characteristics of vehicles within the road network, such as acceleration / deceleration, lane-changing behavior, and following distance. This microscopic dynamic information is crucial for a deeper understanding of traffic flow evolution and for predicting the occurrence and development of traffic congestion—something many traditional methods struggle to achieve.

[0063] Step S20 involves pre-constructing a road network optimization model based on ant colony optimization combined with deep learning. Vehicle dynamic feature information is used as initial pheromone to instill this model. The model simulates vehicle dynamic behavior using a cell transport algorithm, and this behavior serves as heuristic information for the main road segments. Combining the initial pheromone and the heuristic information from the main road segments, a pheromone diffusion model based on ant colony optimization is constructed. This model accurately reflects the traffic operation status and potential of road segments. Compared to simply relying on fixed road attribute information, this heuristic information is more timely and dynamic, providing more valuable guidance for pheromone diffusion in the ant colony optimization algorithm and helping the algorithm search for optimal paths more efficiently.

[0064] Step S30: Load the pheromone diffusion model and solve for the pheromone diffusion function of the model, using pheromone update, vehicle path duration, and vehicle obstacle avoidance count as constraints. When solving the pheromone diffusion function, multiple constraints such as pheromone update, vehicle path duration, and vehicle obstacle avoidance count are considered simultaneously. This allows the obtained cluster solution of the pheromone diffusion function to more comprehensively meet the needs of actual traffic optimization while balancing different traffic objectives.

[0065] Step S40: Obtain the cluster solution of the pheromone diffusion function, and perform decision analysis on the cluster solution of the pheromone diffusion function by combining the dynamic network adjustment algorithm and the multi-objective optimization algorithm to output the traffic signal optimization control strategy and the vehicle optimization path strategy.

[0066] In this embodiment of the invention, a pre-constructed road network optimization model based on ant colony optimization and deep learning is used to comprehensively perceive and analyze the traffic conditions of the entire road network. The full coverage of the road network allows the optimization model to comprehensively consider the mutual influence between various traffic nodes. Furthermore, the introduction of Cellular Transmission (CTM) algorithm to simulate the dynamic behavior of vehicles in the road network enables the optimization model to adjust the guidance strategy according to real-time traffic conditions, better cope with traffic flow fluctuations and emergencies, and combine dynamic network adjustment algorithm and multi-objective optimization algorithm to perform cluster solution decision analysis of pheromone diffusion function. Based on traffic flow data, the pheromone diffusion function and guidance strategy are dynamically adjusted, realizing the coordinated optimization of traffic signal control and vehicle guidance. By outputting traffic signal optimization control strategy and vehicle optimization path strategy, signal timing and guidance path can cooperate with each other to improve the operating efficiency of the traffic system and ensure the accuracy and effectiveness of the guidance strategy.

[0067] This invention provides a method for identifying dynamic characteristic information of vehicles within a road network. Figure 2 A schematic diagram illustrating the implementation process of a method for identifying dynamic characteristic information of vehicles within a road network is shown. The method for identifying dynamic characteristic information of vehicles within a road network specifically includes:

[0068] Step S101: Load real-time traffic flow data of the road network, clean and process outliers in the real-time traffic flow data of the road network to obtain preprocessed real-time traffic flow data of the road network. The real-time traffic flow data of the road network includes, but is not limited to, vehicle identification, traffic volume of main / secondary road segments, vehicle speed, traffic light duration, vehicle initial / end position, and road segments. In implementation, data cleaning tools or data processing libraries in programming languages ​​(such as Python) (such as Pandas) are used to load the data. Then, missing values ​​are filled by simple statistical methods (such as mean, median, etc.). Outliers are identified and processed using box plots or Z-Score methods. Finally, the preprocessed real-time traffic flow data of the road network is obtained.

[0069] Step S102: Based on principal component analysis, real-time traffic flow data of different types of road networks are weighted to obtain real-time weights of data types that represent the importance of the data. The real-time weights of data types are then mapped to a pre-constructed data level evaluation mapping table, and at least one set of real-time vectors of data types are output. The data is standardized using principal component analysis (PCA), and then the covariance matrix and eigenvalues ​​are calculated. The weights of each data type are determined according to the magnitude of the eigenvalues. Using the weights of data types as an index, the real-time weights of data types are mapped to the pre-constructed data level evaluation mapping table. This enables rapid evaluation and classification of data types, which facilitates subsequent data processing and analysis.

[0070] Step S103: Obtain at least one set of real-time vectors of different data types, and concatenate and fuse the at least one set of real-time vectors of different data types to obtain a real-time comprehensive vector representing vehicle dynamic information. During concatenation, the real-time vectors of different data types are arranged and combined according to their time order and spatial location attributes. Then, matrix operations are used to achieve concatenation and fusion to obtain a real-time comprehensive vector. The concatenated real-time comprehensive vector can integrate multi-source data, make full use of the information of each data type, form a more comprehensive and accurate description of vehicle dynamic characteristics, and provide richer data support for subsequent traffic network construction and analysis.

[0071] Step S104: Identify the traffic flow, speed, traffic light duration, and route segments of the main / secondary road segments in the real-time traffic flow data of the road network. Using the complex network method, abstract the road segment intersections as nodes of the traffic network, and abstract the main / secondary road segment traffic as the connecting edges of the traffic network nodes to obtain a dynamic directed traffic network. When constructing the dynamic directed traffic network, a graph theory library (such as NetworkX) can be used to construct the dynamic directed traffic network. The information such as the traffic flow, speed, traffic light duration, and route segments of the main / secondary road segments is stored and managed as network attributes. By abstracting the road segment intersections as nodes of the traffic network and the main / secondary road segment traffic flow as the connecting edges of the traffic network nodes using the complex network method, the complex road network structure can be simplified into a dynamic directed traffic network, which facilitates the analysis and modeling of the dynamic characteristics and topology of traffic flow, and provides a powerful tool for subsequent traffic flow simulation and optimization.

[0072] Step S105: Extract traffic flow, speed data, and traffic light duration of the main / secondary road segments in the dynamic directed traffic network. Normalize the traffic flow, speed data, and traffic light duration of the main / secondary road segments based on the seepage analysis mechanism to obtain a traffic flow matrix, a speed matrix, and a light duration matrix. Superimpose the traffic flow matrix, speed matrix, and light duration matrix to obtain a multi-scale fusion matrix of the main / secondary road segments.

[0073] Step S106: The DDPG algorithm based on deep learning is used to calculate the multi-scale fusion matrix of the main / secondary road segments to obtain the transient seepage index that characterizes the transient congestion level of the main / secondary road segments. When obtaining the transient seepage index, a deep learning framework (such as TensorFlow or PyTorch) is used to build the DDPG algorithm model. The multi-scale fusion matrix is ​​used as input data. The model is trained to learn the characteristics and patterns of traffic flow, and finally the transient seepage index is obtained. The transient seepage index is 0-1.

[0074] Step S107: Using the road segments in the real-time traffic flow data as the correlation, the real-time comprehensive vector is corrected by the transient seepage index to obtain the corrected comprehensive vector. The corrected comprehensive vector is used as the dynamic feature information of vehicles in the road network and output. By using the road segments in the real-time traffic flow data as the correlation and correcting the real-time comprehensive vector by the transient seepage index, the accuracy and timeliness of the vehicle dynamic feature information can be further improved, making it more consistent with the actual situation and providing a more reliable basis for the subsequent traffic control and guidance strategy formulation.

[0075] This invention provides a method for identifying dynamic vehicle feature information within a road network. During vehicle dynamic feature information identification, multi-dimensional data such as vehicle identification, traffic flow, vehicle speed, and traffic light duration are fused into a single-dimensional comprehensive vector. Timestamp alignment and temporal encoding are preserved, reducing computational complexity while maintaining spatiotemporal correlation. A transient seepage index is calculated in real-time based on a deep reinforcement learning algorithm to accurately quantify road segment congestion. Weighted correction of the comprehensive vector using the transient seepage index significantly reduces noise interference and improves feature robustness, thereby enhancing the accuracy and timeliness of vehicle dynamic feature information, making it more realistic, and providing a reliable basis for traffic control and guidance strategy formulation.

[0076] This invention provides a method for pre-constructing a road network optimization model based on ant colony optimization combined with deep learning. Figure 3 The diagram illustrates the implementation flow of a method for pre-constructing a road network optimization model based on ant colony optimization and deep learning. The method specifically includes:

[0077] Step S201: A hybrid architecture of the spatiotemporal prediction model ST-GCN and the sequence prediction model LSTM is used as the road network optimization model. ST-GCN can capture the spatiotemporal correlation between road network topology and traffic flow, while the LSTM model can characterize the dynamics of time series. The combination of the spatiotemporal prediction model ST-GCN and the sequence prediction model LSTM achieves "spatial + temporal" dual-dimensional feature extraction. An autoencoder is introduced before the hybrid architecture, and a deep learning-based feature fusion module is introduced after the hybrid architecture. The autoencoder can process the input data... The system extracts and reduces features to remove noise and redundant information. The feature fusion module further integrates deep learning features and pheromone information from the ant colony algorithm, giving the entire model architecture stronger feature representation capabilities and richer information fusion. The feature fusion module incorporates an ant colony optimization engine, which introduces the ant colony algorithm. Following the feature fusion module is a multi-scale optimization module, which includes a signal optimization layer and a path guidance layer. A classification decision layer is then set after the multi-scale optimization module to output traffic signal optimization control strategies and vehicle path optimization strategies, thus integrating the ant colony algorithm into the deep learning framework. The ant colony algorithm, as a swarm intelligence optimization algorithm that simulates the foraging behavior of ants, can quickly find the optimal or near-optimal solution to a problem through the positive feedback mechanism of pheromones. In road network optimization, this helps to more efficiently search for optimal traffic signal control strategies and vehicle path planning schemes in complex traffic networks.

[0078] Step S202: Capture historical traffic flow data of the road network, analyze the historical traffic flow data of the road network, identify historical dynamic feature information within the road network, and divide the historical dynamic feature information into training set and test set. It should be noted that the method of analyzing historical traffic flow data of the road network is similar to that of analyzing real-time traffic flow data of the road network, and will not be repeated here. The ratio of training set to test set is 4:1.

[0079] Step S203: Define the loss functions and optimizers for the autoencoder, hybrid infrastructure, feature fusion module, and multi-scale optimization module, respectively. The learning rate strategies for the autoencoder, hybrid infrastructure, feature fusion module, and multi-scale optimization module are cosine annealing, learning rate decay, stepwise unfreezing, and adaptive supervision, respectively. Preset the number of training epochs and hyperparameters. The number of training epochs can be 100-120. The autoencoder uses mean squared error (MSE) as the loss function and updates parameters based on the Adam optimizer. The loss function for the hybrid infrastructure can be the cross-entropy loss function, and the optimizer can be the RMSprop optimizer. The loss functions for the feature fusion module and multi-scale optimization module can be the cross-entropy loss function, and the optimizer can be the Momentum optimizer.

[0080] Step S204: The training set is used to pre-train the autoencoder based on unsupervised learning, and the encoded spatiotemporal sequence features are extracted. Gaussian noise is added during training to enhance the robustness of the autoencoder. After pre-training, the parameters of the autoencoder are frozen. The spatiotemporal sequence features extracted by the autoencoder and the training set are fused and input into the hybrid infrastructure. The loss between the predicted output and the true value is calculated. Backpropagation is used to update the weights of the spatiotemporal prediction model and the sequence prediction model. The iteration is repeated until the loss of the hybrid infrastructure converges. Through the pre-training of the autoencoder, the encoded spatiotemporal sequence features are extracted, which can map high-dimensional and complex traffic flow data to a low-dimensional feature space, retain key information in the data, reduce data dimensionality and computational complexity, and provide an efficient and compact feature representation for the subsequent training of the hybrid infrastructure.

[0081] Step S205: Load the spatiotemporal sequence features output by the hybrid infrastructure. Use these features as initial pheromones and introduce a self-supervised reinforcement learning mechanism to jointly train the feature fusion module and the multi-scale optimization module. During the joint training of the feature fusion module and the multi-scale optimization module, the weights and parameters of the feature fusion module and the multi-scale optimization module are dynamically adjusted using an ant colony optimization engine. The converged feature fusion module and multi-scale optimization module are output. After the road network optimization model is trained, the difference between the classification result and the true result is determined using a multi-task joint loss function. The parameters of the road network optimization model are adjusted based on an end-to-end fine-tuning strategy. This process is repeated iteratively until the loss converges, and the converged road network optimization model is output. Using spatiotemporal sequence features as initial pheromones and introducing a self-supervised reinforcement learning mechanism to jointly train the feature fusion module and the multi-scale optimization module enables the fusion and optimization of deep features. The ant colony optimization engine dynamically adjusts the weights and parameters of the feature fusion module and the multi-scale optimization module, further improving the model's ability to model the complex characteristics of traffic flow and its optimization effect, enabling the model to better adapt to different traffic scenarios and dynamically changing traffic demands.

[0082] The multi-task joint loss function of the road network optimization model is expressed as:

[0083] (1)

[0084] (2)

[0085] (3)

[0086] (4)

[0087] in, This represents the multi-task joint loss function of the road network optimization model. These represent signal optimization loss, path-induced loss, and signal reconstruction loss, respectively. These are the signal control weight, path-inducing weight, and signal reconstruction weight, respectively. In this embodiment, they can be 0.4, 0.5, and 0.2, respectively. As a penalty item for energy consumption, These are the changes in average travel time, the average congestion points along the route, and the changes in speed. These represent the weighting coefficients for average travel time, average congestion points along the route, and speed change, respectively, and the sum of these weighting coefficients is 1. These are the predicted path distribution and the actual path distribution, respectively. These represent the deviation penalty coefficient and the delay deviation, respectively. These are the initial pheromones input and the heuristic information for the main road segments of the road network, respectively. Indicates time consistency weight, The time derivatives of the initial pheromones and the heuristic information of the main road segments of the road network are respectively used. In this embodiment of the invention, a multi-task joint loss function is used to determine the difference between the classification result and the true result, and the parameters of the road network optimization model are adjusted based on an end-to-end fine-tuning strategy. This approach can simultaneously consider the objectives of two tasks: traffic signal optimization control strategy and vehicle path optimization strategy, achieving collaborative optimization of multi-task learning. This collaborative optimization method can improve the overall performance of the model on multiple tasks and avoid the problem that optimizing a single task may lead to a decline in the performance of other tasks.

[0088] Step S206: Obtain the test set, use the test set as input, execute the pre-trained road network optimization model, obtain the test results, and calculate the prediction accuracy of the test results;

[0089] Step S207: Determine whether the prediction accuracy meets the preset test accuracy threshold, which can be 0.85-0.9.

[0090] Step S208: If the prediction accuracy meets the preset test accuracy threshold, output the converged road network optimization model.

[0091] If the prediction accuracy does not meet the preset test accuracy threshold, the process returns to step S204 to continue iterative training of the model, thus forming a closed-loop mechanism for continuous improvement. This helps to continuously optimize model performance, improve its prediction accuracy and optimization effect on road network traffic flow, and ensure that the model can adapt to constantly changing traffic conditions and complex and ever-changing road network environments.

[0092] This invention provides a road network optimization model and its construction method. By constructing a hybrid infrastructure including a spatiotemporal prediction model ST-GCN and a sequence prediction model LSTM, and introducing autoencoders and feature fusion modules before and after, the model can fully mine and integrate the spatiotemporal and sequence features in road network traffic flow data. The effective feature extraction and dimensionality reduction of the autoencoder, along with the deep fusion processing of the feature fusion module, enables the model to more accurately capture the complex patterns and trends of traffic flow, thereby significantly improving the prediction accuracy of key traffic parameters such as traffic flow and vehicle speed. Furthermore, an ant colony optimization engine is introduced, combining ant colony algorithms with deep learning. Utilizing the global search capability and positive feedback mechanism of ant colony algorithms, the optimal traffic signal control strategy and vehicle path planning scheme are efficiently searched in complex road network environments. The road network optimization model can organically combine traffic signal optimization control with vehicle path guidance, achieving coordinated operation between the two. By optimizing traffic signal timing and vehicle travel paths, the utilization efficiency of road resources is improved, vehicle queuing time and delays are reduced, and energy consumption and environmental pollution are decreased.

[0093] This invention provides a method for simulating vehicle dynamic behavior based on a cell transmission algorithm. Figure 4 This diagram illustrates the implementation flow of a method for simulating vehicle dynamic behavior based on a cell transport algorithm. The method specifically includes:

[0094] Step S301: Obtain vehicle dynamic feature information. The autoencoder encodes and processes the vehicle dynamic feature information, and outputs the encoded spatiotemporal sequence features. The autoencoder extracts features with strong spatiotemporal correlation (such as congestion propagation patterns) through unsupervised learning, so that the feature dimension of the vehicle dynamic feature information is expanded from the original 64 dimensions to 256 dimensions.

[0095] Step S302: The hybrid infrastructure performs spatiotemporal prediction and sequence prediction processing on the spatiotemporal sequence features to obtain spatiotemporal sequence features containing spatiotemporal prediction results and sequence prediction results. The spatiotemporal sequence features are used as initial pheromones. The feature fusion module dynamically corrects the initial pheromones based on the ant colony optimization engine to obtain pheromone correction pheromones. The feature fusion module dynamically corrects the initial pheromones based on the ant colony optimization engine. Combined with the global search capability and positive feedback mechanism of the ant colony algorithm, it can dynamically adjust the distribution and intensity of pheromones according to real-time traffic conditions, so that the model can better adapt to changes in traffic flow, further improve the accuracy of prediction and optimization effect, and provide more realistic guidance information for vehicle dynamic behavior simulation.

[0096] Step S303: Obtain the dynamic directed traffic network, use information modifiers as heuristic information for the dynamic directed traffic network, initialize the dynamic directed traffic network, divide the road segments in the dynamic directed traffic network into cell units of equal length, update the maximum flow, upstream flow, downstream flow, and transmission rate of the cell units based on the heuristic information of the dynamic directed traffic network, and update parameters such as the maximum flow, upstream flow, downstream flow, and transmission rate of the cell units according to the heuristic information of the dynamic directed traffic network. This enables the cell transmission model to more flexibly adapt to the traffic flow characteristics of different road segments, accurately reflect the capacity and traffic congestion of each road segment, and enhance the model's ability to capture dynamic changes in traffic flow.

[0097] Step S304: Load the updated cell unit. Simulate the cell unit based on the preset cell unit time step, vehicle transmission calculation parameters, traffic density threshold, and number of simulations to obtain vehicle dynamic behavior. Use the vehicle dynamic behavior as heuristic information for the main road segments of the road network. During cell unit simulation, a pheromone diffusion model based on the ant colony algorithm is constructed by combining the initial pheromone and the heuristic information of the main road segments of the road network. The pheromone diffusion model is then optimized and output as an optimized pheromone diffusion model. In the cell unit simulation process, a pheromone diffusion model based on the ant colony algorithm is constructed by combining the initial pheromone and the heuristic information of the main road segments of the road network and optimized. This can provide real-time feedback on changes in traffic flow and dynamically adjust the vehicle's driving path and behavior strategy through the pheromone diffusion mechanism, thereby achieving self-organized optimization of traffic flow, effectively alleviating traffic congestion, and improving the traffic efficiency of the road network.

[0098] Step S305: Load the pheromone diffusion model, and solve the pheromone diffusion function of the pheromone diffusion model with constraints such as updating pheromone, vehicle path duration, and vehicle obstacle avoidance times, to obtain the cluster solution of the pheromone diffusion function.

[0099] In this embodiment of the invention, when simulating vehicle dynamic behavior based on the cell transmission algorithm, the high-dimensional vehicle dynamic feature data is reduced in dimensionality through the encoding process, making the data easier to process and analyze, while retaining the core information in the data. This helps to improve the training speed and prediction accuracy of the subsequent model. The feature fusion module dynamically corrects the initial pheromones based on the ant colony optimization engine. Combined with the global search capability and positive feedback mechanism of the ant colony algorithm, it can dynamically adjust the distribution and intensity of pheromones according to real-time traffic conditions, enabling the model to better adapt to changes in traffic flow, further improving the accuracy of prediction and optimization effect, and providing more realistic guidance information for vehicle dynamic behavior simulation.

[0100] This invention provides a method for decision analysis of clustered solutions to pheromone diffusion functions by combining dynamic network adjustment algorithms and multi-objective optimization algorithms. Figure 5This diagram illustrates the implementation flow of a method combining dynamic network adjustment algorithm and multi-objective optimization algorithm for cluster solution decision analysis of pheromone diffusion function. The method specifically includes:

[0101] Step S401: Load the pheromone diffusion function of the pheromone diffusion model. With the objectives of balancing traffic flow in the road network and optimizing the phase time and green light time of traffic signals, dynamically balance the traffic flow, phase time, and green light time based on random bubble sort to obtain the cluster solution of the pheromone diffusion function. The pheromone diffusion function is expressed as:

[0102] (5)

[0103] in, Represents the pheromone diffusion function. Indicates the pheromone update factor. These represent the congestion attenuation factor and the path congestion duration, respectively. For the number of simulations per cell unit, These represent the total number of congested points and the average number of congested points along the route. Indicates real-time traffic intensity;

[0104] Step S402: Update the parameters of the pheromone diffusion model based on the dynamic network adjustment algorithm, and use the fitness function to calculate the individual optimal position and global optimal position of the vehicle during the iterative update based on the ant colony algorithm.

[0105] Step S403: Combine the dynamic pheromone mapping mechanism to map vehicle dynamic behavior into ant colony behavior parameters. Based on the multi-objective optimization algorithm, dynamically fuse the cluster solution of pheromone diffusion function and ant colony behavior parameters to obtain at least one set of dynamic fused solutions. Sort and filter the dynamic fused solutions using the TOPSIS method to determine the final road network optimization strategy. The road network optimization strategy includes traffic signal optimization control strategy and vehicle optimization path strategy. Output the traffic signal optimization control strategy and vehicle optimization path strategy.

[0106] In this embodiment, the objectives are to balance traffic flow in the road network and optimize the phase time and green light time of traffic signals. Dynamic balancing is achieved based on random bubble sort. This allows for real-time adjustment of the weights and priorities of each objective according to different traffic conditions and demands, avoiding traffic flow imbalances that may result from optimizing a single objective. This achieves dynamic balance of road network traffic flow and refined optimization of traffic signals. By calculating the individual optimal position and the global optimal position, a combination of global search capability and local mining capability is achieved. This not only finds the optimal path and signal configuration globally but also fully considers local traffic characteristics and details, improving the feasibility and effectiveness of the optimization scheme. The TOPSIS method is used to sort and filter the dynamic fusion solutions. This is a scientific and objective multi-attribute decision-making method that comprehensively considers the distance between each fusion solution and the ideal solution and the negative ideal solution, accurately assesses the quality of each solution, and ensures that the final determined road network optimization strategy has the best comprehensive performance, providing a reliable basis for traffic management decisions.

[0107] This invention provides a traffic control and guidance system based on ant colony algorithm. Figure 6 A schematic diagram of a traffic control and guidance system based on ant colony algorithm is shown. This system specifically includes:

[0108] The traffic flow acquisition module 100 acquires real-time traffic flow data of the road network based on computer vision, analyzes the real-time traffic flow data of the road network, and identifies the dynamic characteristic information of vehicles in the road network.

[0109] The vehicle simulation module 200 pre-constructs a road network optimization model based on ant colony algorithm combined with deep learning. It uses vehicle dynamic feature information as initial pheromone to give to the road network optimization model, simulates vehicle dynamic behavior based on cell transmission algorithm, uses vehicle dynamic behavior as heuristic information for the main road segments of the road network, and constructs a pheromone diffusion model based on ant colony algorithm by combining the initial pheromone and the heuristic information of the main road segments of the road network.

[0110] The pheromone diffusion module 300 is used to load the pheromone diffusion model, and with the constraints of updating pheromones, vehicle path duration, and vehicle obstacle avoidance times, solve the pheromone diffusion function of the pheromone diffusion model to obtain the cluster solution of the pheromone diffusion function.

[0111] The road network optimization module 400 is used to obtain the cluster solution of the pheromone diffusion function, and to perform decision analysis on the cluster solution of the pheromone diffusion function by combining the dynamic network adjustment algorithm and the multi-objective optimization algorithm, and output the traffic signal optimization control strategy and the vehicle optimization path strategy.

[0112] In this embodiment, the traffic flow acquisition module 100 includes:

[0113] Preprocessing unit 110 is used to load real-time traffic flow data of the road network, clean the real-time traffic flow data of the road network and handle outliers to obtain preprocessed real-time traffic flow data of the road network.

[0114] The weight mapping unit 120 assigns weights to real-time traffic flow data of different types of road networks based on principal component analysis, obtains real-time weights of data types that characterize the importance of the data, maps the real-time weights of data types to a pre-built data level evaluation mapping table, and outputs at least one set of real-time vectors of data types.

[0115] The traffic network construction unit 130 is used to acquire at least one set of real-time vectors of data types, splice and fuse at least one set of real-time vectors of data types to obtain a real-time comprehensive vector representing vehicle dynamic information, identify the traffic flow, vehicle speed, traffic light duration and route segments of the main / secondary road segments in the real-time traffic flow data of the road network, abstract the road segment intersections as nodes of the traffic network according to the complex network method, and abstract the main / secondary road segment vehicles as the connecting edges of the traffic network nodes to obtain a dynamic directed traffic network.

[0116] The multi-scale fusion unit 140 is used to extract traffic flow, speed data and traffic light duration of the main / secondary road segments in the dynamic directed traffic network. Based on the seepage analysis mechanism, the traffic flow, speed data and traffic light duration of the main / secondary road segments are normalized to obtain the traffic flow matrix, speed matrix and light duration matrix. The traffic flow matrix, speed matrix and light duration matrix are superimposed to obtain the multi-scale fusion matrix of the main / secondary road segments.

[0117] The feature information output unit 150 uses the deep learning DDPG algorithm to calculate the multi-scale fusion matrix of the main / secondary road segments, and obtains the transient seepage index that characterizes the transient congestion level of the main / secondary road segments. It uses the road segments in the real-time traffic flow data as the correlation, and corrects the real-time comprehensive vector through the transient seepage index to obtain the corrected comprehensive vector. The corrected comprehensive vector is used as the dynamic feature information of vehicles in the road network and output.

[0118] It should be noted that the traffic control and guidance system based on ant colony algorithm provided in this embodiment of the invention corresponds to the traffic control and guidance method based on ant colony algorithm described above. The explanations, examples, and beneficial effects of the relevant content can be referred to the corresponding content in the traffic control and guidance method based on ant colony algorithm, and will not be repeated here.

[0119] In summary, this invention provides a traffic control and guidance system and method based on ant colony optimization. In the embodiments of this invention, by pre-constructing a road network optimization model based on ant colony optimization combined with deep learning, the traffic situation of the entire road network can be comprehensively perceived and analyzed. The full coverage of the road network allows the road network optimization model to comprehensively consider the mutual influence between various traffic nodes. Furthermore, the introduction of Cell Transmission (CTM) algorithm to simulate the dynamic behavior of vehicles in the road network enables the road network optimization model to adjust the guidance strategy according to real-time traffic conditions, better cope with traffic flow fluctuations and sudden events, and combine dynamic network adjustment algorithm and multi-objective optimization algorithm to perform cluster solution decision analysis of pheromone diffusion function. Based on traffic flow data, the pheromone diffusion function and guidance strategy are dynamically adjusted, realizing the coordinated optimization of traffic signal control and vehicle guidance. By outputting traffic signal optimization control strategy and vehicle optimization path strategy, signal timing and guidance path can cooperate with each other to improve the operating efficiency of the traffic system and ensure the accuracy and effectiveness of the guidance strategy.

[0120] It should be noted that, for the sake of simplicity, the foregoing embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to the present invention. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0121] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on these embodiments, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art can still combine, add, delete, or otherwise adjust the features of the various embodiments of the present invention according to the circumstances without conflict or creative effort, thereby obtaining different technical solutions that do not fundamentally depart from the concept of the present invention. These technical solutions also fall within the scope of protection of the present invention.

Claims

1. A traffic control and guidance method based on ant colony algorithm, characterized in that, The method includes: Real-time traffic flow data of road network is acquired based on computer vision, and the real-time traffic flow data of road network is analyzed to identify the dynamic characteristics of vehicles in the road network. A road network optimization model based on ant colony algorithm combined with deep learning is pre-constructed. Vehicle dynamic feature information is used as the initial pheromone to give the road network optimization model. The dynamic behavior of vehicles is simulated based on the cell transmission algorithm. The dynamic behavior of vehicles is used as the heuristic information of the main road segments of the road network. The pheromone diffusion model based on ant colony algorithm is constructed by combining the initial pheromone and the heuristic information of the main road segments of the road network. Load the pheromone diffusion model, and with the pheromone update, vehicle path duration, and vehicle obstacle avoidance count as constraints, solve the pheromone diffusion function of the pheromone diffusion model to obtain the cluster solution of the pheromone diffusion function. Obtain cluster solutions of pheromone diffusion function, combine dynamic network adjustment algorithm and multi-objective optimization algorithm to make decision analysis on cluster solutions of pheromone diffusion function, and output traffic signal optimization control strategy and vehicle optimization path strategy; The method for pre-constructing a road network optimization model based on ant colony optimization combined with deep learning includes: A hybrid infrastructure for the road network optimization model is adopted, using the spatiotemporal prediction model ST-GCN and the sequence prediction model LSTM. An autoencoder is introduced before the hybrid infrastructure, and a feature fusion module based on deep learning is introduced after the hybrid infrastructure. The feature fusion module has a built-in ant colony optimization engine, which incorporates the ant colony algorithm. A multi-scale optimization module is introduced after the feature fusion module. The multi-scale optimization module includes a signal optimization layer and a path guidance layer. A classification decision layer is set after the multi-scale optimization module. The classification decision layer is used to output traffic signal optimization control strategy and vehicle path optimization strategy. Capture historical traffic flow data of the road network, analyze the historical traffic flow data of the road network, identify historical dynamic feature information within the road network, and divide the historical dynamic feature information into training set and test set; Define the loss functions and optimizers for the autoencoder, hybrid infrastructure, feature fusion module, and multi-scale optimization module respectively, and preset the training epochs and hyperparameters; The training set is used to pre-train the autoencoder based on unsupervised learning, extracting the spatiotemporal sequence features after encoding. Gaussian noise is added during training to enhance the robustness of the autoencoder. After pre-training, the parameters of the autoencoder are frozen. The spatiotemporal sequence features extracted by the autoencoder are fused with the training set and input into the hybrid infrastructure. The loss between the predicted output and the true value is calculated. Backpropagation is used to update the weights of the spatiotemporal prediction model and the sequence prediction model. The iteration is repeated until the loss of the hybrid infrastructure converges. The spatiotemporal sequence features output by the hybrid infrastructure are loaded and used as initial pheromones. A self-supervised reinforcement learning mechanism is introduced to jointly train the feature fusion module and the multi-scale optimization module. During the joint training of the feature fusion module and the multi-scale optimization module, the weights and parameters of the feature fusion module and the multi-scale optimization module are dynamically adjusted through the ant colony optimization engine. The converged feature fusion module and multi-scale optimization module are output. After the road network optimization model is trained, the difference between the classification result and the true result is determined by the multi-task joint loss function. The parameters of the road network optimization model are adjusted based on the end-to-end fine-tuning strategy. The iteration is repeated until the loss converges and the converged road network optimization model is output. Obtain a test set, use the test set as input, execute the pre-trained road network optimization model, obtain test results, calculate the prediction accuracy of the test results, determine whether the prediction accuracy meets the preset test accuracy threshold, and if the prediction accuracy meets the preset test accuracy threshold, output the converged road network optimization model. The method for simulating vehicle dynamic behavior based on the cell transport algorithm includes: The system acquires vehicle dynamic feature information, processes the vehicle dynamic feature information using an autoencoder, and outputs the encoded spatiotemporal sequence features. The hybrid infrastructure performs spatiotemporal prediction and sequence prediction processing on spatiotemporal sequence features to obtain spatiotemporal sequence features containing spatiotemporal prediction results and sequence prediction results. The spatiotemporal sequence features are used as initial pheromones, and the feature fusion module dynamically corrects the initial pheromones based on the ant colony optimization engine to obtain pheromone corrections. The dynamic directed traffic network is obtained, and the information modifier is used as the heuristic information of the dynamic directed traffic network. The dynamic directed traffic network is initialized, and the road segments in the dynamic directed traffic network are divided into cell units of the same length. The maximum flow, upstream flow, downstream flow and transmission rate of the cell units are updated based on the heuristic information of the dynamic directed traffic network. The updated cell unit is loaded and simulated based on the preset cell unit time step, vehicle transmission calculation parameters, traffic density threshold, and number of simulations to obtain vehicle dynamic behavior. The vehicle dynamic behavior is used as heuristic information for the main road segments of the road network. During the cell unit simulation, a pheromone diffusion model based on the ant colony algorithm is constructed by combining the initial pheromone and the heuristic information of the main road segments of the road network. The pheromone diffusion model is then optimized and the optimized pheromone diffusion model is output.

2. The traffic control and guidance method based on ant colony algorithm as described in claim 1, characterized in that: The method for identifying dynamic characteristic information of vehicles within a road network includes: Load real-time traffic flow data of the road network, clean the real-time traffic flow data of the road network and handle outliers to obtain pre-processed real-time traffic flow data of the road network. The real-time traffic flow data of the road network includes vehicle identification, traffic volume of main road segment and secondary road segment, vehicle speed, traffic light duration, vehicle initial position and ending position, and road segment passed through. Based on the principal component analysis method, real-time traffic flow data of different types of road networks are weighted to obtain real-time weights of data types that represent the importance of data. The real-time weights of data types are then mapped to a pre-constructed data level evaluation mapping table, and at least one set of real-time vectors of data types are output. Obtain at least one set of real-time vectors of different data types, and concatenate and fuse the at least one set of real-time vectors of different data types to obtain a real-time comprehensive vector representing vehicle dynamic information. Identify the traffic volume, speed, traffic light duration, and route segments of the main and secondary road segments in the real-time traffic flow data of the road network. Using the complex network method, road segment intersections are abstracted as nodes of the traffic network, and vehicles on the main and secondary road segments are abstracted as connecting edges of the traffic network nodes, thus obtaining a dynamic directed traffic network. Traffic flow, speed, and traffic light duration data of main and secondary road segments are extracted from the dynamic directed traffic network. Based on the seepage analysis mechanism, the traffic flow, speed, and traffic light duration data of the main and secondary road segments are normalized to obtain traffic flow matrix, speed matrix, and traffic light duration matrix. The traffic flow matrix, speed matrix, and traffic light duration matrix are superimposed to obtain the multi-scale fusion matrix of the main and secondary road segments. The DDPG algorithm, a deep learning algorithm, is used to calculate the multi-scale fusion matrix of the main road section and the secondary road section, and a transient seepage index is obtained to characterize the transient congestion level of the main road section and the secondary road section. By associating the road segments in the real-time traffic flow data with transient seepage indices, the real-time comprehensive vector is corrected to obtain the corrected comprehensive vector. The corrected comprehensive vector is then used as the dynamic characteristic information of vehicles within the road network and output.

3. The traffic control and guidance method based on ant colony algorithm as described in claim 1, characterized in that: The method for cluster solution decision analysis of pheromone diffusion function by combining dynamic network adjustment algorithm and multi-objective optimization algorithm includes: The pheromone diffusion function of the pheromone diffusion model is loaded. With the goal of balancing traffic flow in the road network and optimizing the phase time and green light time of traffic signals, the pheromone diffusion function cluster solution is obtained by dynamically balancing the traffic flow in the road network, optimizing the phase time and green light time of traffic signals based on random bubble sort. The parameters of the pheromone diffusion model are updated based on the dynamic network adjustment algorithm, and the fitness function is used to calculate the individual optimal position and the global optimal position of the vehicle during the iterative update based on the ant colony algorithm.

4. The traffic control and guidance method based on ant colony algorithm as described in claim 3, characterized in that: The method for cluster solution decision analysis of pheromone diffusion function combining dynamic network adjustment algorithm and multi-objective optimization algorithm further includes: By combining a dynamic pheromone mapping mechanism to map vehicle dynamic behavior into ant colony behavior parameters, and using a multi-objective optimization algorithm to dynamically fuse the cluster solution of the pheromone diffusion function and the ant colony behavior parameters, at least one set of dynamic fused solutions is obtained. The dynamic fused solutions are sorted and filtered by the TOPSIS method to determine the final road network optimization strategy. The road network optimization strategy includes a traffic signal optimization control strategy and a vehicle optimization path strategy, and outputs the traffic signal optimization control strategy and the vehicle optimization path strategy.

5. A traffic control and guidance system based on ant colony algorithm, used to implement the traffic control and guidance method based on ant colony algorithm as described in any one of claims 1-4, characterized in that: The traffic control and guidance system based on the ant colony algorithm includes: The traffic flow acquisition module acquires real-time traffic flow data of the road network based on computer vision, analyzes the real-time traffic flow data of the road network, and identifies the dynamic characteristics of vehicles within the road network. The vehicle simulation module pre-constructs a road network optimization model based on ant colony algorithm combined with deep learning. Vehicle dynamic feature information is used as initial pheromone to give the road network optimization model. The module simulates vehicle dynamic behavior based on cell transmission algorithm and uses vehicle dynamic behavior as heuristic information for the main road segments of the road network. The module combines the initial pheromone and the heuristic information of the main road segments of the road network to construct a pheromone diffusion model based on ant colony algorithm. The pheromone diffusion module is used to load the pheromone diffusion model, and with the constraints of updating pheromone, vehicle path duration, and vehicle obstacle avoidance times, solve the pheromone diffusion function of the pheromone diffusion model to obtain the cluster solution of the pheromone diffusion function. The road network optimization module is used to obtain cluster solutions of the pheromone diffusion function, and to conduct decision analysis on the cluster solutions of the pheromone diffusion function by combining dynamic network adjustment algorithm and multi-objective optimization algorithm, and output traffic signal optimization control strategy and vehicle optimization path strategy.

6. The traffic control and guidance system based on ant colony algorithm as described in claim 5, characterized in that: The traffic flow acquisition module includes: The preprocessing unit is used to load real-time traffic flow data of the road network, clean the real-time traffic flow data of the road network and handle outliers to obtain preprocessed real-time traffic flow data of the road network. The weight mapping unit assigns weights to real-time traffic flow data of different types of road networks based on principal component analysis, obtains real-time weights of data types that characterize the importance of the data, maps the real-time weights of data types to a pre-built data level evaluation mapping table, and outputs at least one set of real-time vectors of data types. The traffic network construction unit is used to acquire at least one set of real-time vectors of data types, splice and fuse at least one set of real-time vectors of data types to obtain a real-time comprehensive vector representing vehicle dynamic information, identify the traffic flow, vehicle speed, traffic light duration and route segments of the main road segment and the secondary road segment in the real-time traffic flow data of the road network, abstract the road segment intersections as nodes of the traffic network according to the complex network method, and abstract the main road segment and the secondary road segment vehicles as the connecting edges of the traffic network nodes to obtain a dynamic directed traffic network. The multi-scale fusion unit is used to extract traffic flow, speed data, and traffic light duration of the main road segment and the secondary road segment in the dynamic directed traffic network. Based on the seepage analysis mechanism, the traffic flow, speed data, and traffic light duration of the main road segment and the secondary road segment are normalized to obtain the traffic flow matrix, the speed matrix, and the traffic light duration matrix. The traffic flow matrix, the speed matrix, and the traffic light duration matrix are superimposed to obtain the multi-scale fusion matrix of the main road segment and the secondary road segment. The feature information output unit uses the deep learning DDPG algorithm to calculate the multi-scale fusion matrix of the main road segment and the secondary road segment, and obtains the transient seepage index that characterizes the transient congestion degree of the main road segment and the secondary road segment. The transit road segments in the real-time traffic flow data are used as the correlation, and the real-time comprehensive vector is corrected by the transient seepage index to obtain the corrected comprehensive vector. The corrected comprehensive vector is used as the dynamic feature information of vehicles in the road network and output.

Images