Sensitivity and calculation integrated management and control system and method for highway confluence area

Abstract

The invention discloses a system and method for integrated management and control of communication, sensing and calculation of a highway confluence area, and the system is characterized in that a vehicle terminal and a roadside terminal are connected with an edge calculation unit, and transmit sensing data to the edge calculation unit; the vehicle terminal and the roadside terminal are loaded with SNN models and execute VAI and IAI tasks respectively, the edge computing unit is loaded with a DNN model to execute EAI tasks, and a dynamic distributed computing power network is jointly constructed and formed; the edge calculation unit carries out coordination and cooperates with the vehicle terminal and the roadside terminal to carry out distributed iterative training of the edge federated learning model until convergence; and the edge calculation unit determines a traffic operation state, a vehicle control strategy and safety early warning information according to the data, and feeds back the information to the vehicle terminal and the roadside terminal. Through the technical scheme of the invention, the data processing efficiency and decision accuracy are remarkably improved, intelligent prediction and management of traffic congestion are realized, and smoother and safer travel experience is provided for drivers.

Description

Technical Field

[0001] The present invention relates to the technical field of road traffic control systems, and in particular to a synergy-computing integrated control system for a merging area of ​​a highway and a synergy-computing integrated control method for a merging area of ​​a highway. Background Art

[0002] As a key area where vehicles converge, the traffic flow in the merging area of ​​a highway is complex and changeable, with a high risk of traffic accidents and easy to cause traffic congestion. In the traditional highway merging area control system, the perception, communication, and computing functions are often performed independently and decentralized. The perception function mainly relies on radar, video and other equipment installed on the roadside to collect information such as vehicle flow, speed, and vehicle distance. The perception equipment can often only provide relatively basic data, and the accuracy and resolution are limited by the hardware itself. The communication function is to use the communication network (wired communication, wireless communication) to transmit the perception data to the management center or edge computing unit. The communication process will cause delays or network congestion. The computing function is to analyze and process the received data in the management center or edge computing unit to evaluate the traffic operation status of the merging area and formulate corresponding control strategies.

[0003] With the development of mobile communication and artificial intelligence technologies, communication networks and AI are constantly merging. Communication networks are gradually evolving from central intelligence to edge intelligence, giving rise to a new network paradigm with "edge intelligence" capabilities, aiming to provide ubiquitous real-time intelligent services. Edge intelligent networks have expanded application scenarios such as smart highways through the collaboration and interaction of communication, perception, and computing. The data-driven highway merge area control scenario not only puts forward more extreme performance requirements for communication and computing, but also puts forward high-precision and high-resolution requirements for perception. In terms of perception, higher-precision data such as vehicle location, speed, and distance are required, while high-resolution image and video data provide the basis for subsequent analysis and decision-making. In terms of communication, it is necessary to build an edge intelligent network that can support random access of roadside nodes, and the network has low-latency information perception, transmission, and processing capabilities. In terms of computing, it is necessary to use technologies such as artificial intelligence and machine learning to deeply mine and analyze traffic data and formulate corresponding control strategies.

[0004] In order to meet the needs of high-precision, high-resolution perception data in highway merging areas, as well as to achieve low-latency data transmission and efficient computing processing, an innovative control system and device are urgently needed. At the same time, the system also needs to build a stable, reliable, and low-latency edge intelligent communication network to support random access to roadside nodes and high-speed data transmission. Summary of the invention

[0005] In response to the above problems, the present invention provides a system and method for integrated control of highway merging areas by integrating advanced technologies of communication, perception, computing and control, building a real-time interactive and collaborative management network for vehicles, roads and traffic operation systems, and significantly improving data processing efficiency and decision-making accuracy by introducing a dynamic distributed computing power network architecture and a refined multi-level AI task classification strategy. At the same time, the variable speed limit model and federated learning technology are used to achieve intelligent prediction and management of traffic congestion, optimize driving safety, and transform the highway merging area control system from a model of "first perception, then communication and finally calculation" to a model of "distributed iteration of model parameters based on perception data", providing drivers with a smoother and safer travel experience, and promoting the intelligent transportation system to move towards a more intelligent and efficient direction.

[0006] To achieve the above-mentioned object, the present invention provides a synergy-computing integrated management and control system for a merging area of ​​a highway, comprising: a vehicle terminal, a roadside terminal and an edge computing unit;

[0007] The vehicle terminal is communicatively connected to the edge computing unit via a roadside unit RSU, and the roadside terminal is communicatively connected to the edge computing unit;

[0008] The vehicle terminal sends the vehicle perception data acquired by the vehicle perception device to the edge computing unit through the RSU, and the roadside terminal sends the roadside perception data acquired by the roadside perception device to the edge computing unit;

[0009] The vehicle terminal and the roadside terminal are loaded with SNN models, respectively execute VAI tasks and IAI tasks, and send the task calculation results to the edge computing unit respectively. The edge computing unit is loaded with DNN model to execute EAI tasks, and calculate the information uploaded by the vehicle terminal and the roadside terminal. The vehicle terminal, the roadside terminal and the edge computing unit jointly construct a dynamic distributed computing network;

[0010] The edge computing unit performs distributed machine learning with the vehicle terminal and the roadside terminal based on an edge federated learning model, and the edge computing unit coordinates and cooperates with the vehicle terminal and the roadside terminal to perform iterative training until the edge federated learning model converges;

[0011] The edge computing unit performs fusion calculations based on the acquired perception data, determines the traffic operation status, vehicle control strategy and safety warning information based on the preset merging area section of the highway, and feeds back to the vehicle terminal and the roadside terminal.

[0012] In the above technical solution, preferably, the vehicle perception device includes a vehicle-mounted sensor, a vehicle-mounted radar and a vehicle-mounted camera, and the roadside perception device includes a roadside radar, a roadside camera and a weather detector;

[0013] The roadside terminal also includes a variable information sign, which receives and publishes the safety warning information pushed by the edge computing unit.

[0014] In the above technical solution, preferably, the roadside terminals are distributed in upstream and downstream sections and ramp sections of the main line merging point of a preset merging area of ​​the expressway.

[0015] In the above technical solution, preferably, the vehicle terminal performs VAI tasks based on the SNN model with the support of local computing resources on the vehicle side, and processes the vehicle perception data acquired by the vehicle perception device;

[0016] The roadside terminal performs the IAI task based on the SNN model with the support of local computing resources at the roadside, and processes the roadside sensing data acquired by the roadside sensing device;

[0017] The edge computing unit performs EAI tasks based on the DNN model under the support of edge computing resources, and performs fusion processing on the perception data sent by the vehicle terminal and the roadside terminal;

[0018] Wherein, the SNN model and the DNN model are pre-trained offline using massive historical data.

[0019] In the above technical solution, preferably, the vehicle terminal, the roadside terminal and the edge computing unit use OTA technology to remotely upgrade the model;

[0020] The vehicle terminal, the roadside terminal and the edge computing unit perform distributed machine learning based on the edge federated learning model, and the learning process includes:

[0021] The edge computing unit broadcasts the model parameters of the edge federated learning model to the vehicle terminal and the roadside terminal through a downlink, so that the vehicle terminal and the roadside terminal are synchronized and initialize corresponding local models;

[0022] The vehicle terminal and the roadside terminal train and update local model parameters based on local data sets;

[0023] The vehicle terminal and the roadside terminal send the updated local model parameters to the edge computing unit via an uplink for model parameter aggregation;

[0024] The training is iterated until the edge federated learning model converges.

[0025] The present invention also proposes a method for integrated control of a merging area of ​​a highway using a sensory computing system, which is used in a merging area of ​​a highway using a sensory computing system as disclosed in any one of the above technical solutions, comprising:

[0026] Acquire vehicle perception data through a vehicle terminal, and execute a VAI task to process the vehicle perception data;

[0027] Acquire roadside sensing data through a roadside terminal, and execute an IAI task to process the roadside sensing data;

[0028] The processed vehicle perception data and the roadside perception data are sent to the edge computing unit, and the perception data are fused by executing the EAI task;

[0029] The variable speed limit rules are determined based on traffic congestion, and the congestion prediction model of the main line of the merging area based on KNN-LSTM is trained according to the historical traffic data of the main line section of the preset merging area of ​​the expressway;

[0030] According to the fusion processing result of the perception data by the edge computing unit, the congestion level of the main line section of the merging area is predicted by the merging area main line congestion prediction model, and the vehicle control strategy and safety warning information of the merging area section are obtained according to the variable speed limit rule;

[0031] The vehicle control strategy is fed back to the vehicle terminal through the RSU for vehicle control, and the safety warning information is fed back to the roadside terminal for warning information release.

[0032] In the above technical solution, preferably, the variable speed limit rule is determined based on traffic congestion, and a congestion prediction model for the main line of the merging area based on KNN-LSTM is trained according to historical traffic data of the main line section of the merging area preset in the expressway. The specific process includes:

[0033] Classify the traffic congestion of expressway sections and determine the warning speed limit value for each level of congestion;

[0034] According to the historical average speed data of the main line section of the preset merging area, the k-means clustering algorithm is used to perform clustering to obtain the speed range of each level of congestion in the current section, wherein the number of cluster centers set is equal to the number of levels of traffic congestion;

[0035] The historical average speed data set is divided into a training set and a test set, and training is performed based on the recurrent neural network LSTM to obtain a merging area mainline congestion prediction model based on KNN-LSTM, so that the merging area mainline congestion prediction model can predict the congestion level according to real-time perception data.

[0036] In the above technical solution, preferably, according to the fusion processing result of the perception data by the edge computing unit, the congestion level of the main line section of the corresponding merging area is predicted by the merging area main line congestion prediction model, and the vehicle control strategy and safety warning information of the merging area section are obtained according to the variable speed limit rule. The specific process includes:

[0037] The edge computing unit calculates the average vehicle speed of the current road section according to the sensing data, inputs the average vehicle speed into the merging area main line congestion prediction model, and obtains the congestion level of the current merging area main line section;

[0038] According to the variable speed limit rule, a warning speed limit value under the current congestion level of the merging area section is obtained;

[0039] According to the traffic operation status and surrounding vehicle information around each vehicle, combined with the warning speed limit value, a vehicle control strategy and safety warning information for each vehicle are obtained;

[0040] Among them, the vehicle control strategy includes adjusting lane change and acceleration and deceleration strategies, and the safety warning information includes obstacle reminders ahead of the road, vehicle close-range danger warnings, rear vehicle overtaking warnings, side vehicle collision warnings and speed limit reminders.

[0041] In the above technical solution, preferably, the integrated control method of the synaptic computing in the highway merging area further includes:

[0042] Using massive historical data to perform offline pre-training on the SNN model and the DNN model;

[0043] Deploy the trained SNN model and the DNN model to the vehicle terminal, the roadside terminal and the edge computing unit respectively;

[0044] The specific process of acquiring vehicle perception data through the vehicle terminal and executing the VAI task to process the vehicle perception data, acquiring roadside perception data through the roadside terminal and executing the IAI task to process the roadside perception data, and sending the processed vehicle perception data and the roadside perception data to the edge computing unit, and fusing the perception data by executing the EAI task includes:

[0045] The vehicle terminal performs the VAI task based on the SNN model with the support of local computing resources on the vehicle side, and processes the vehicle perception data;

[0046] The roadside terminal performs the IAI task based on the SNN model with the support of local computing resources at the roadside to process the roadside perception data;

[0047] The edge computing unit executes the EAI task based on the DNN model with the support of edge computing resources, and performs fusion processing on the perception data sent by the vehicle terminal and the roadside terminal.

[0048] In the above technical solution, preferably, OTA technology is used to remotely upgrade the models of the vehicle terminal, the roadside terminal and the edge computing unit;

[0049] Distributed machine learning is performed on the vehicle terminal, the roadside terminal, and the edge computing unit based on the edge federated learning model, and the learning process includes:

[0050] The edge computing unit broadcasts the model parameters of the edge federated learning model to the vehicle terminal and the roadside terminal through a downlink, so that the vehicle terminal and the roadside terminal are synchronized and initialize corresponding local models;

[0051] The vehicle terminal and the roadside terminal train and update local model parameters based on local data sets;

[0052] The vehicle terminal and the roadside terminal send the updated local model parameters to the edge computing unit via an uplink for model parameter aggregation;

[0053] The training is iterated until the edge federated learning model converges.

[0054] Compared with the prior art, the beneficial effects of the present invention are: by integrating advanced technologies of communication, perception, computing and control, a real-time interactive and collaborative management network for vehicle, road and traffic operation system is constructed, and by introducing a dynamic distributed computing power network architecture and a refined multi-level AI task classification strategy, data processing efficiency and decision-making accuracy are significantly improved. At the same time, by using a variable speed limit model and federated learning technology, intelligent prediction and management of traffic congestion are realized, driving safety is optimized, and the highway merging area control system is transformed from a model of "first perception, then communication and finally calculation" to a model of "distributed iteration of model parameters based on perception data", providing drivers with a smoother and safer travel experience, and promoting the intelligent transportation system to move towards a more intelligent and efficient direction. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A schematic diagram of the architecture of a synaptic computing integrated control system for a merging area of ​​a highway disclosed in an embodiment of the present invention;

[0056] Figure 2 A schematic diagram of the perception-computation integration process of a synergy-computation integrated management and control system for a merging area of ​​a highway disclosed in an embodiment of the present invention;

[0057] Figure 3A schematic diagram of a communication integrated control hierarchical model of a communication sensing and computing integrated control system for a merging area of ​​a highway disclosed in an embodiment of the present invention;

[0058] Figure 4 A schematic diagram of the learning process of an edge federation machine learning model disclosed in an embodiment of the present invention;

[0059] Figure 5 A schematic diagram of the division of the interweaving area of ​​the main line and ramp merging section of an expressway disclosed in an embodiment of the present invention;

[0060] Figure 6 A schematic diagram of the facility layout of a synaptic computing integrated control system for a merging area of ​​a highway disclosed in an embodiment of the present invention;

[0061] Figure 7 A schematic diagram of the prediction process of a merging area mainline congestion prediction model based on KNN-LSTM disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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.

[0063] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0064] like Figure 1 and Figure 2 As shown, a synergy-computing integrated management and control system for a merging area of ​​a highway provided by the present invention includes: a vehicle terminal, a roadside terminal and an edge computing unit;

[0065] The vehicle terminal is connected to the edge computing unit through the roadside unit RSU, and the roadside terminal is connected to the edge computing unit;

[0066] The vehicle terminal sends the vehicle perception data acquired by the vehicle perception device to the edge computing unit through the RSU, and the roadside terminal sends the roadside perception data acquired by the roadside perception device to the edge computing unit;

[0067] The vehicle terminal and the roadside terminal are equipped with SNN models, which respectively execute VAI tasks and IAI tasks, and send the task calculation results to the edge computing unit. The edge computing unit is equipped with DNN models to execute EAI tasks and calculate the information uploaded by the vehicle terminal and the roadside terminal. The vehicle terminal, the roadside terminal and the edge computing unit jointly build a dynamic distributed computing network.

[0068] The edge computing unit, vehicle terminal, and roadside terminal perform distributed machine learning based on the edge federated learning model. The edge computing unit coordinates and cooperates with the vehicle terminal and roadside terminal to perform iterative training until the edge federated learning model converges.

[0069] The edge computing unit performs fusion calculations based on the acquired perception data, determines the traffic operation status, vehicle control strategy and safety warning information based on the preset merging area section of the highway, and feeds back to the vehicle terminal and roadside terminal.

[0070] In this implementation, by integrating advanced technologies of communication, perception, computing and control, a real-time interactive and collaborative management network for vehicles, roads and traffic operation systems is constructed. By introducing a dynamic distributed computing power network architecture and a refined multi-level AI task classification strategy, data processing efficiency and decision-making accuracy are significantly improved. At the same time, by using a variable speed limit model and federated learning technology, intelligent prediction and management of traffic congestion are achieved, driving safety is optimized, and the highway merging area control system is transformed from a "first perception, then communication and finally calculation" model to a "distributed iteration of model parameters based on perception data" model, providing drivers with a smoother and safer travel experience, and promoting the intelligent transportation system to move towards a more intelligent and efficient direction.

[0071] Specifically, the above-mentioned integrated control system has the following characteristics:

[0072] (1) Integration of perception and computing. The system realizes an integrated process from data collection to processing. The vehicle side and the roadside side pre-process the original perception information before uploading it, and the edge computing unit is responsible for centralized processing and decision-making, which significantly improves the efficiency and accuracy of data processing.

[0073] (2) Distributed computing power network: A dynamic distributed computing power network for highway merging areas is built to achieve on-demand allocation and efficient utilization of computing power resources, ensuring the real-time and high efficiency of computing tasks.

[0074] (3) Multi-level AI task classification: According to the different task processing locations, AI tasks are divided into three categories: VAI, IAI, and EAI, which correspond to the vehicle side, roadside side, and edge computing unit, respectively. This effectively utilizes the computing power of equipment at all levels and improves the overall performance of the system.

[0075] (4) Variable speed limit model: variable speed limit rules are designed based on traffic congestion, and the KNN-LSTM model is used to predict road congestion, realizing intelligent management of main line speed limits in merging areas, effectively alleviating traffic congestion and improving driving safety.

[0076] Furthermore, through integrated design, the present invention deeply integrates perception, computing and communication technologies for the first time, constructs an integrated sensing and computing management and control system for highway merging areas, and realizes comprehensive perception and intelligent processing of traffic information; secondly, by constructing a dynamic distributed computing power network, it innovatively proposes a dynamic distributed computing power network for highway merging areas, realizes flexible scheduling and efficient utilization of computing power resources, and provides strong support for large-scale data processing and real-time decision-making; finally, the application of multi-level AI task classification and federated learning models not only optimizes the system resource allocation, but also improves the model iteration speed and machine learning capabilities, laying the foundation for the sustainable development of intelligent transportation systems.

[0077] Among them, the integrated control system of the merging area of ​​the expressway consists of 1 control center, L roadside units (RSU) and edge computing units, M multi-functional terminals and N targets to be sensed. The vehicle terminal and the roadside terminal use radar, cameras and other equipment to collect various data such as vehicle operation status, road status and weather status on the road in real time. The RSU receives the perception data of the vehicle terminal and sends the data to the edge computing unit. The edge computing unit or the control center aggregates the vehicle perception data of the vehicle terminal and the roadside perception data of the roadside terminal, and formulates the control strategy of the merging area using the built-in algorithm model. In the integrated control system of the merging area of ​​the expressway, the servers of the vehicle terminal, the roadside terminal and the edge computing unit build a dynamic distributed computing network for the merging area of ​​the expressway, providing computing power support for computing tasks such as multi-dimensional perception information fusion, feature extraction and autonomous decision-making of the vehicle terminal, the roadside terminal and the edge computing unit.

[0078] Specifically, in the integrated control system of the merging area of ​​the expressway, the RSU can provide vehicles within its coverage area with vehicle-road collaborative services such as cooperative driving, beyond-visual-range perception, and collision warning, and provide transit communication between the vehicle terminal and the edge computing unit. The edge computing unit needs to use artificial intelligence AI algorithms to integrate and process various types of perception information perceived by the vehicle terminal and the roadside terminal. The vehicle terminal processes the original perception information obtained by on-board sensors such as radar and video, and uploads the processed result information to the RSU, which is then uploaded to the edge computing unit by the RSU. Similarly, the roadside terminal processes the original perception information collected by roadside perception devices such as radar and video, and uploads the processed result information to the edge computing unit. The edge computing unit server can obtain vehicle perception information and roadside perception information in real time, so that it can make processing decisions on perception information in a centralized control manner. AI tasks are tasks that are processed by pre-trained AI models, and outputs are obtained by executing reasoning. In the integrated control system of the merging area of ​​the expressway, AI tasks are represented as the fusion, dimensionality reduction, and recognition of various types of perception information.

[0079] Specifically, according to the location where the AI ​​tasks are processed (vehicle terminal, roadside terminal or edge computing unit), AI tasks can be further classified into VAI (Vehicle Artificial Intelligence) tasks, IAI (Infrastructure Artificial Intelligence) tasks and EAI (Edge Artificial Intelligence) tasks.

[0080] (1) VAI tasks:

[0081] At time k, vehicle i has computing tasks that need to be processed by AI algorithms. The amount of computing power consumed by these algorithms depends on the size of the video and radar perception information of vehicle i. The VAI task of vehicle i is processed locally, and the vehicle uploads the result information to the RSU.

[0082] (2) IAI Mission

[0083] At time k, the roadside terminal j has computing tasks that need to be processed by AI algorithms. The amount of computing power consumed by these algorithms depends on the amount of information collected by the video and radar equipment deployed by the roadside terminal j. The IAI tasks of the roadside terminal j are processed locally, and the roadside terminal j uploads the result information to the edge computing unit.

[0084] (3)EAI Tasks

[0085] At time k, edge computing unit j has computing tasks that need to be processed by AI algorithms. The computing power consumed by this algorithm depends on the size of the perception information uploaded by the vehicle terminal and the roadside terminal (original perception information or locally processed information).

[0086] In the above embodiment, preferably, the vehicle perception device includes a vehicle-mounted sensor, a vehicle-mounted radar and a vehicle-mounted camera, and the roadside perception device includes a roadside radar, a roadside camera and a weather detector;

[0087] The roadside terminal also includes a variable information sign, which receives and publishes safety warning information pushed by the edge computing unit.

[0088] In order for vehicles to safely and smoothly pass through the main line and ramp merging sections, they need to obtain real-time information on the surrounding traffic operation status, including vehicle speed, traffic events, vehicle queues, etc.

[0089] In order to adjust vehicle control strategies such as lane changing, acceleration and deceleration, the vehicle needs to obtain information such as the position and speed of surrounding vehicles in real time;

[0090] In order to avoid traffic accidents, vehicles need to obtain real-time safety warning information such as obstacle reminders ahead on the road, vehicle close-range danger warnings, rear vehicle overtaking, and side vehicle collision warnings.

[0091] Specifically, in the sections where the main line and the ramp merge, the road environment changes and the risk of traffic operation safety increases. Roadside smart facilities such as perception, communication, and control are deployed, and a vehicle-road collaborative system is built. Visibility and road condition detection equipment are deployed to perceive the traffic meteorological environment in real time and issue meteorological warning information to ensure driving safety in the sections where the main line and the ramp merge.

[0092] Among them, the sensing facilities include:

[0093] (1) Traffic status perception: traffic flow, average vehicle speed, occupancy rate, etc.

[0094] (2) Vehicle operation status perception: single vehicle speed, vehicle trajectory, abnormal vehicle acceleration and deceleration, etc.;

[0095] (3) Traffic event perception: queue length, traffic congestion, traffic accidents, etc.;

[0096] (4) AI computing power (SNN).

[0097] Among them, communication facilities include:

[0098] (1) Wired communication facilities support the transmission of large-bandwidth data such as images and videos;

[0099] (2) Wireless communication facilities RSU.

[0100] Among them, edge computing units and control facilities include:

[0101] (1) Edge computing unit: Accesses information pushed by vehicle terminals and roadside terminal sensing facilities, analyzes and processes the information, generates safety warning information such as dynamic speed limits and traffic incidents (AI computing capability, DNN), and pushes it to roadside RSUs and variable information signs;

[0102] (2) Roadside RSU: Receives the processed or original perception information from the vehicle terminal, receives the safety warning information pushed by the edge computing terminal, and pushes it to the on-board OBU to provide voice / image information to the driver;

[0103] (3) Variable information sign: Receives safety warning information pushed by the edge computing unit and publishes it to the driver.

[0104] In the above-mentioned embodiment, preferably, the roadside terminals are distributed in upstream and downstream sections and ramp sections of the main line merging point of a preset merging area of ​​the expressway.

[0105] Specifically, Figure 5 As shown in the figure, due to the differences in the design of each ramp, the number and location of the roadside facilities and meteorological environment detection facilities in the actual project need to be determined according to the actual situation of each main line and ramp merging section. In order to ensure the driving safety of the main line and ramp merging area, the basic layout principle is: no blind spot coverage.

[0106] like Figure 6 As shown, the basic layout plan is:

[0107] Main line: A set of integrated sensing and computing facilities are deployed on the upstream and downstream sections of the confluence point, including video, radar, edge computing, RSU and other equipment;

[0108] Ramp: deploy a set of integrated sensing and computing facilities, including video, radar, edge computing, RSU and other equipment; if the ramp has a large curvature, or is long and wide, and one set of equipment cannot cover it, then additional corresponding equipment must be added as needed.

[0109] In the above implementation, preferably, the vehicle terminal performs the VAI task based on the SNN model with the support of the local computing resources on the vehicle side, and processes the vehicle perception data acquired by the vehicle perception device;

[0110] The roadside terminal performs IAI tasks based on the SNN model with the support of local computing resources at the roadside, and processes the roadside perception data obtained by the roadside perception equipment;

[0111] The edge computing unit performs EAI tasks based on the DNN model with the support of edge computing resources, and integrates the perception data sent by the vehicle terminal and the roadside terminal;

[0112] Among them, the SNN model and DNN model use massive historical data for offline pre-training.

[0113] In the above implementation, preferably, the vehicle terminal, the roadside terminal and the edge computing unit use OTA technology to remotely upgrade the model;

[0114] The vehicle terminal, roadside terminal and edge computing unit perform distributed machine learning based on the edge federated learning model. The learning process includes:

[0115] The edge computing unit broadcasts the model parameters of the edge federated learning model to the vehicle terminal and the roadside terminal through the downlink, so that the vehicle terminal and the roadside terminal are synchronized and initialize the corresponding local model;

[0116] The vehicle terminal and the roadside terminal train and update the local model parameters based on the local data set;

[0117] The vehicle terminal and the roadside terminal send the updated local model parameters to the edge computing unit via the uplink for model parameter aggregation;

[0118] Iterate training until the edge federated learning model converges.

[0119] Specifically, the above-mentioned Tonggansuan integrated management and control system also has the characteristics of an efficient model algorithm upgrade strategy: OTA (Over-the-Air Technology) technology is used to realize remote upgrades of global, edge and local models, and edge computing units and multi-functional terminals implement distributed machine learning based on the federated learning model, which not only ensures data privacy and security, but also improves the model iteration speed and machine learning capabilities.

[0120] like Figure 3 As shown in the figure, in this technical system, the algorithm model is divided into three levels: global model, edge model, and local model. Both edge devices (vehicle terminals, roadside terminals, and edge computing units) and center servers use OTA technology to remotely upgrade the model algorithm. The model algorithms of the center and edge are completed by downloading and installing new program packages. However, in the core link of algorithm upgrades - data iteration and machine learning, there are significant differences between the edge and the center.

[0121] The central server has far more computing power than the edge. While processing computing tasks, the central server can allocate a large amount of computing power (for example, 50% computing power) for machine learning, which makes the data iteration speed of the central end fast and the machine learning ability strong, and the global model can be quickly upgraded at the central end. The computing power of edge devices is mainly used to perform edge computing tasks and perception tasks, and only limited computing power (for example, 10% to 20% computing power) is left for machine learning. The limited computing power limits its ability to perceive, calculate, and learn and train at the same time. In addition, machine learning of edge devices usually takes a long time and the learning speed is slow. For the evolution and update of the model algorithm supported by the central server, the upgraded model is pushed to the edge computing unit through OTA technology.

[0122] like Figure 4 As shown in the figure, the edge computing unit, vehicle terminal, and roadside terminal implement distributed machine learning with privacy protection based on the edge federated learning model. The edge federated learning model uses the computing power of multiple smart terminals, coordinated by the edge computing unit, to collaboratively train the edge model in an iterative manner. The process of the edge federated learning process includes the following 3 steps.

[0123] Step 1: The edge computing unit broadcasts the edge model parameters to the terminal through the downlink so that all terminals participating in federated learning can synchronize and initialize their local models.

[0124] Step 2: Each terminal participating in federated learning trains and updates its local model parameters based on its own dataset.

[0125] Step 3: The terminal sends the updated local model parameters to the edge computing unit of the base station via the uplink for edge model parameter aggregation. The iteration stops when the edge model converges.

[0126] According to the integrated control system of the synaptic computing of the highway merging area disclosed in the above-mentioned embodiment, the perception process and the computing process are integrated in the system architecture. On the basis of the AI ​​task processing of the terminal, the data is transmitted to the edge computing unit through the communication unit for fusion computing, forming a multi-level AI task classification processing method, and constructing a dynamic distributed computing power network to achieve on-demand allocation and efficient utilization of computing power resources. Among them, based on the calculation data, the control strategy calculation and analysis can be performed according to the built-in algorithm model, and the corresponding control strategy information can be fed back to the vehicle terminal and the roadside terminal to realize the vehicle control of the highway merging area.

[0127] The present invention further proposes a method for integrated control of a merging area of ​​a highway using a sensory computing system, which is used in a merging area of ​​a highway using a sensory computing system as disclosed in any one of the above embodiments, comprising:

[0128] Acquire vehicle perception data through the vehicle terminal and execute VAI tasks to process the vehicle perception data;

[0129] Obtain roadside perception data through roadside terminals and execute IAI tasks to process the roadside perception data;

[0130] The processed vehicle perception data and roadside perception data are sent to the edge computing unit, and the perception data is fused and processed by executing the EAI task;

[0131] The variable speed limit rules are determined based on traffic congestion, and the congestion prediction model of the main line of the merging area based on KNN-LSTM is trained according to the historical traffic data of the main line section of the preset merging area of ​​the expressway;

[0132] According to the fusion processing results of the edge computing unit on the perception data, the congestion level of the main line section of the merging area is predicted by the merging area main line congestion prediction model, and the vehicle control strategy and safety warning information of the merging area section are obtained according to the variable speed limit rule;

[0133] The vehicle control strategy is fed back to the vehicle terminal through the RSU for vehicle management and control, and the safety warning information is fed back to the roadside terminal for warning information release.

[0134] In this implementation, by integrating advanced technologies of communication, perception, computing and control, a real-time interactive and collaborative management network for vehicles, roads and traffic operation systems is constructed. By introducing a dynamic distributed computing power network architecture and a refined multi-level AI task classification strategy, data processing efficiency and decision-making accuracy are significantly improved. At the same time, by using a variable speed limit model and federated learning technology, intelligent prediction and management of traffic congestion are achieved, driving safety is optimized, and the highway merging area control system is transformed from a "first perception, then communication and finally calculation" model to a "distributed iteration of model parameters based on perception data" model, providing drivers with a smoother and safer travel experience, and promoting the intelligent transportation system to move towards a more intelligent and efficient direction.

[0135] During the implementation process, based on the perception data collected in real time by the vehicle terminal and the roadside terminal, and with the computing power support of the dynamic distributed computing network composed of the vehicle terminal, the roadside terminal and the edge computing unit, the congestion of the corresponding road section is predicted and analyzed through the KNN-LSTM-based merging area mainline congestion prediction model, and the vehicle speed limit warning value of the current lane is comprehensively calculated according to the predetermined variable speed limit rules.

[0136] Based on the above calculated data, the vehicle in the current lane can be controlled to realize the control of the unmanned vehicle. Warning information can also be issued to the vehicles in the current lane to remind the driver to drive safely. Warning information can also be issued through the roadside terminal of the current section of the road to remind the driver to drive safely.

[0137] In the above implementation, preferably, a variable speed limit rule is determined based on traffic congestion, and a congestion prediction model for the main line of the merging area based on KNN-LSTM is trained based on historical traffic data of the main line section of the merging area preset by the expressway. The specific process includes:

[0138] Classify the traffic congestion of expressway sections and determine the warning speed limit value for each level of congestion;

[0139] According to the historical average speed data of the main line section of the preset merging area, the k-means clustering algorithm is used to cluster and obtain the speed range of each level of congestion in the current section, where the number of cluster centers set is equal to the number of levels of traffic congestion;

[0140] The historical average speed dataset is divided into a training set and a test set, and training is performed based on the recurrent neural network LSTM to obtain a merging area mainline congestion prediction model based on KNN-LSTM. This enables the merging area mainline congestion prediction model to predict the congestion level based on real-time perception data.

[0141] Specifically, the control purpose of the variable speed limit rule is to reduce the discrete degree of vehicle speed, control vehicle speed in advance, ensure safety while alleviating traffic congestion, and reduce the probability of secondary accidents. The speed limit triggered by the system is determined according to the traffic congestion, and different levels of traffic congestion correspond to different optimal speed limit values.

[0142] Rules: Variable speed limit rules based on traffic congestion:

[0143] If C=N,then S limit =V congestion

[0144] Among them, C is the traffic congestion, N is the congestion value, S limit (km / h) is the warning speed limit value, V congestion

[0145] (km / h) is the warning speed limit value under different levels of congestion.

[0146] The classification of traffic congestion status is based on micro parameters and traffic data, among which vehicle speed is the most commonly used traffic parameter for evaluating traffic status. Based on the vehicle speed as the congestion classification standard, the congestion of expressway sections is divided into five levels. On the basis of the traffic congestion classification, the variable speed limit rules based on traffic congestion are determined, see Table 1.

[0147] Table 1 Variable speed limit rules based on traffic congestion

[0148]

[0149] During the implementation process, the k-means clustering algorithm was used. Since the road congestion level is divided into five levels: smooth, basically smooth, general, congested, and blocked, the cluster center of the clustering algorithm was set to 5.

[0150] like Figure 7 As shown in the figure, the model is trained by inputting the historical average speed. After the model training, the historical average speed is divided into 5 categories. The upper and lower limits of each category correspond to a standard interval for dividing the congestion level. The LSTM recurrent neural network divides the historical average speed data set into a training set and a test set. During training, the number of network layers and the number of neurons in each layer are adjusted to make the network's prediction accuracy as high as possible.

[0151] In the above implementation, preferably, according to the fusion processing result of the edge computing unit on the perception data, the congestion level of the main line section of the corresponding merging area is predicted by the merging area main line congestion prediction model, and the vehicle control strategy and safety warning information of the merging area section are obtained according to the variable speed limit rule. The specific process includes:

[0152] The edge computing unit calculates the average vehicle speed of the current road section based on the perception data, inputs the average vehicle speed into the congestion prediction model of the main line of the merging area, and obtains the congestion level of the main line section of the current merging area;

[0153] According to the variable speed limit rule, the warning speed limit value under the current congestion level of the merging area section is obtained;

[0154] According to the traffic operation status and surrounding vehicle information around each vehicle, combined with the warning speed limit value, the vehicle control strategy and safety warning information for each vehicle are obtained;

[0155] Among them, the vehicle control strategy includes adjusting the lane change and acceleration and deceleration strategies, and the safety warning information includes obstacle reminders ahead of the road, vehicle close-range danger warnings, rear vehicle overtaking warnings, side vehicle collision warnings and speed limit reminders.

[0156] In the above implementation, preferably, the integrated control method of the expressway merging area by means of telepathy and computing further includes:

[0157] Use massive historical data to pre-train SNN and DNN models offline;

[0158] The trained SNN model and DNN model are deployed to vehicle terminals, roadside terminals and edge computing units respectively.

[0159] On this basis, the vehicle perception data is obtained through the vehicle terminal, and the VAI task is executed to process the vehicle perception data, the roadside perception data is obtained through the roadside terminal, and the IAI task is executed to process the roadside perception data, and the processed vehicle perception data and roadside perception data are sent to the edge computing unit, and the perception data is fused by executing the EAI task. The specific process includes:

[0160] The vehicle terminal performs VAI tasks based on the SNN model with the support of local computing resources on the vehicle side and processes the vehicle perception data;

[0161] The roadside terminal performs IAI tasks based on the SNN model with the support of local computing resources at the roadside and processes the roadside perception data;

[0162] With the support of edge computing resources, the edge computing unit performs EAI tasks based on the DNN model and integrates the perception data sent by the vehicle terminal and the roadside terminal.

[0163] In the above implementation, preferably, OTA technology is used to remotely upgrade the models of the vehicle terminal, the roadside terminal and the edge computing unit;

[0164] Based on the edge federated learning model, distributed machine learning is performed on vehicle terminals, roadside terminals, and edge computing units. The learning process includes:

[0165] The edge computing unit broadcasts the model parameters of the edge federated learning model to the vehicle terminal and the roadside terminal through the downlink, so that the vehicle terminal and the roadside terminal are synchronized and initialize the corresponding local model;

[0166] The vehicle terminal and the roadside terminal train and update the local model parameters based on the local data set;

[0167] The vehicle terminal and the roadside terminal send the updated local model parameters to the edge computing unit via the uplink for model parameter aggregation;

[0168] Iterate training until the edge federated learning model converges.

[0169] According to the integrated control method of the expressway merging area disclosed in the above-mentioned implementation mode, during the implementation process, each step is implemented based on the integrated control system of the expressway merging area disclosed in the above-mentioned implementation mode. During the implementation process, the deployment and implementation methods of each module of the system disclosed in the above-mentioned implementation mode can be referred to for operation, which will not be repeated here.

[0170] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A synergy-computing integrated control system for highway merging areas, characterized in that: include: Vehicle terminals, roadside terminals and edge computing units; The vehicle terminal is communicatively connected to the edge computing unit via a roadside unit RSU, and the roadside terminal is communicatively connected to the edge computing unit; The vehicle terminal sends the vehicle perception data acquired by the vehicle perception device to the edge computing unit through the RSU, and the roadside terminal sends the roadside perception data acquired by the roadside perception device to the edge computing unit; The vehicle terminal and the roadside terminal are loaded with SNN models, respectively execute VAI tasks and IAI tasks, and send the task calculation results to the edge computing unit respectively. The edge computing unit is loaded with DNN model to execute EAI tasks, and calculate the information uploaded by the vehicle terminal and the roadside terminal. The vehicle terminal, the roadside terminal and the edge computing unit jointly construct a dynamic distributed computing network; The edge computing unit performs distributed machine learning with the vehicle terminal and the roadside terminal based on an edge federated learning model, and the edge computing unit coordinates and cooperates with the vehicle terminal and the roadside terminal to perform iterative training until the edge federated learning model converges; The edge computing unit performs fusion calculations based on the acquired perception data, determines the traffic operation status, vehicle control strategy and safety warning information based on the preset merging area section of the highway, and feeds back to the vehicle terminal and the roadside terminal.

2. The integrated control system of expressway merging area according to claim 1 is characterized in that: The vehicle perception device includes a vehicle-mounted sensor, a vehicle-mounted radar and a vehicle-mounted camera, and the roadside perception device includes a roadside radar, a roadside camera and a weather detector; The roadside terminal also includes a variable information sign, which receives and publishes the safety warning information pushed by the edge computing unit.

3. The integrated control system of expressway merging area according to claim 1 is characterized in that: The roadside terminals are distributed in upstream and downstream sections and ramp sections of the main line merging point of a preset merging area of ​​the expressway.

4. The integrated control system of expressway merging area according to claim 1 is characterized in that: The vehicle terminal performs VAI tasks based on the SNN model with the support of local computing resources on the vehicle side, and processes the vehicle perception data acquired by the vehicle perception device; The roadside terminal performs the IAI task based on the SNN model with the support of local computing resources at the roadside, and processes the roadside sensing data acquired by the roadside sensing device; The edge computing unit performs EAI tasks based on the DNN model under the support of edge computing resources, and performs fusion processing on the perception data sent by the vehicle terminal and the roadside terminal; The SNN model and the DNN model are pre-trained offline using massive historical data.

5. The integrated control system of expressway merging area according to claim 1 is characterized in that: The vehicle terminal, the roadside terminal and the edge computing unit use OTA technology to remotely upgrade the model; The vehicle terminal, the roadside terminal and the edge computing unit perform distributed machine learning based on the edge federated learning model, and the learning process includes: The edge computing unit broadcasts the model parameters of the edge federated learning model to the vehicle terminal and the roadside terminal through a downlink, so that the vehicle terminal and the roadside terminal are synchronized and initialize corresponding local models; The vehicle terminal and the roadside terminal train and update local model parameters based on local data sets; The vehicle terminal and the roadside terminal send the updated local model parameters to the edge computing unit via an uplink for model parameter aggregation; The training is iterated until the edge federated learning model converges.

6. A method for integrated control of expressway merging areas by using synaptic computing, characterized in that: The integrated control system for a merging area of ​​a highway according to any one of claims 1 to 5 comprises: Acquire vehicle perception data through a vehicle terminal, and execute a VAI task to process the vehicle perception data; Acquire roadside sensing data through a roadside terminal, and execute an IAI task to process the roadside sensing data; The processed vehicle perception data and the roadside perception data are sent to the edge computing unit, and the perception data are fused by executing the EAI task; The variable speed limit rules are determined based on traffic congestion, and the congestion prediction model of the main line of the merging area based on KNN-LSTM is trained according to the historical traffic data of the main line section of the preset merging area of ​​the expressway; According to the fusion processing result of the perception data by the edge computing unit, the congestion level of the main line section of the merging area is predicted by the merging area main line congestion prediction model, and the vehicle control strategy and safety warning information of the merging area section are obtained according to the variable speed limit rule; The vehicle control strategy is fed back to the vehicle terminal through the RSU for vehicle control, and the safety warning information is fed back to the roadside terminal for warning information release.

7. The integrated control method of expressway merging area based on synaptic computing according to claim 6 is characterized in that: The variable speed limit rule is determined based on traffic congestion, and a congestion prediction model for the main line of the merging area based on KNN-LSTM is trained according to historical traffic data of the main line section of the merging area preset by the expressway. The specific process includes: Classify the traffic congestion of expressway sections and determine the warning speed limit value for each level of congestion; According to the historical average speed data of the main line section of the preset merging area, the k-means clustering algorithm is used to perform clustering to obtain the speed range of each level of congestion in the current section, wherein the number of cluster centers set is equal to the number of levels of traffic congestion; The historical average speed data set is divided into a training set and a test set, and training is performed based on the recurrent neural network LSTM to obtain a merging area mainline congestion prediction model based on KNN-LSTM, so that the merging area mainline congestion prediction model can predict the congestion level according to real-time perception data.

8. The integrated control method of expressway merging area based on synaptic computing according to claim 7 is characterized in that: According to the fusion processing result of the edge computing unit on the perception data, the congestion level of the main line section of the merging area is predicted by the merging area main line congestion prediction model, and the vehicle control strategy and safety warning information of the merging area section are obtained according to the variable speed limit rule. The specific process includes: The edge computing unit calculates the average vehicle speed of the current road section according to the sensing data, inputs the average vehicle speed into the merging area main line congestion prediction model, and obtains the congestion level of the current merging area main line section; According to the variable speed limit rule, a warning speed limit value under the current congestion level of the merging area section is obtained; According to the traffic operation status and surrounding vehicle information around each vehicle, combined with the warning speed limit value, a vehicle control strategy and safety warning information for each vehicle are obtained; Among them, the vehicle control strategy includes adjusting lane change and acceleration and deceleration strategies, and the safety warning information includes obstacle reminders ahead of the road, vehicle close-range danger warnings, rear vehicle overtaking warnings, side vehicle collision warnings and speed limit reminders.

9. The integrated control method of expressway merging area based on synaptic computing according to claim 6 is characterized in that: Also includes: Using massive historical data to perform offline pre-training on the SNN model and the DNN model; Deploy the trained SNN model and the DNN model to the vehicle terminal, the roadside terminal and the edge computing unit respectively; The specific process of acquiring vehicle perception data through the vehicle terminal and executing the VAI task to process the vehicle perception data, acquiring roadside perception data through the roadside terminal and executing the IAI task to process the roadside perception data, and sending the processed vehicle perception data and the roadside perception data to the edge computing unit, and fusing the perception data by executing the EAI task includes: The vehicle terminal performs the VAI task based on the SNN model with the support of local computing resources on the vehicle side, and processes the vehicle perception data; The roadside terminal performs the IAI task based on the SNN model with the support of local computing resources at the roadside to process the roadside perception data; The edge computing unit executes the EAI task based on the DNN model with the support of edge computing resources, and performs fusion processing on the perception data sent by the vehicle terminal and the roadside terminal.

10. The integrated control method of expressway merging area based on synaptic computing according to claim 9, characterized in that: Using OTA technology to remotely upgrade the models of the vehicle terminal, the roadside terminal, and the edge computing unit; Distributed machine learning is performed on the vehicle terminal, the roadside terminal, and the edge computing unit based on the edge federated learning model, and the learning process includes: The edge computing unit broadcasts the model parameters of the edge federated learning model to the vehicle terminal and the roadside terminal through a downlink, so that the vehicle terminal and the roadside terminal are synchronized and initialize corresponding local models; The vehicle terminal and the roadside terminal train and update local model parameters based on local data sets; The vehicle terminal and the roadside terminal send the updated local model parameters to the edge computing unit via an uplink for model parameter aggregation; The training is iterated until the edge federated learning model converges.

Images