Expressway hard shoulder dynamic opening method and system based on NFGD model and simulation platform

Abstract

The invention discloses an expressway hard shoulder dynamic opening method and system based on an NFGD model and a simulation platform. Comprising the following steps: collecting and adopting a dynamic confidence mechanism to carry out weighted fusion on multi-source real-time sensing data to obtain fused traffic state characteristics; a hard road shoulder dynamic control decision is made based on an NFGD model comprising a fuzzy-neural hybrid controller, a multi-target genetic algorithm controller and a reinforcement learning controller; and real-time interaction with a simulation platform is realized, and control instruction issuing, feedback acquisition and online strategy evaluation are realized. Compared with a traditional method based on a fixed empirical threshold value, the fuzzy-neural hybrid controller, the multi-target genetic algorithm controller and the reinforcement learning controller are fused, self-adaptive modeling and open discrimination are achieved, and decision-making precision and scene adaptability are improved. And meanwhile, by combining with a prediction adjustment feedback type step length control structure, simulation time drift is effectively inhibited, the timeliness of a control strategy is enhanced, and the method has relatively high practical application value.

Description

Technical Field

[0001] The present invention relates to the fields of traffic management and intelligent transportation systems, and in particular to a method and system for dynamically opening hard shoulders of highways based on an NFGD model and a simulation platform. Background Art

[0002] With the continuous growth of highway traffic, traditional fixed hard shoulder management strategies are struggling to meet dynamic traffic demands. Existing technologies primarily rely on single sensors to collect static data of limited dimensions and lack the ability to deeply integrate and analyze multi-source, heterogeneous data in real time, resulting in inaccurate and ineffective traffic state assessments. Furthermore, existing hard shoulder opening strategies are often based on manual rules or static thresholds, failing to achieve high-precision synchronization with microscopic traffic simulations (such as SUMO). The lack of a closed-loop feedback mechanism between simulation results and real-time sensor data limits the dynamic adaptability of control strategies. At the decision-making level, traditional methods often only consider a single optimization objective (such as traffic efficiency) and are unable to intelligently coordinate conflicting multiple objectives such as efficiency, safety, and disturbance minimization. Furthermore, they lack the self-learning capabilities to adapt to complex traffic scenarios (such as inclement weather or mixed traffic flows). Furthermore, the existing systems' sensing, simulation, and control modules lack synergy and inconsistent data exchange standards, making closed-loop optimization from state perception to policy execution difficult. This results in delayed evaluation of hard shoulder opening effectiveness and an inability to continuously iterate and optimize control parameters using historical data. Therefore, there is an urgent need for a dynamic opening solution for highway hard shoulders that integrates multi-source data fusion, high-precision simulation synchronization, and intelligent decision-making to improve road network traffic efficiency and safety management. Summary of the Invention

[0003] Purpose of the invention: To overcome the defects of existing highway hard shoulder opening strategies, such as reliance on manual rules, delayed response, and weak adaptability, the present invention proposes a method and system for dynamic opening of highway hard shoulders based on the NFGD model and simulation platform. By integrating the three modules of simulation, perception, and control, the present invention independently designs a number of intelligent control algorithms and feedback mechanisms, which can realize dynamic opening, real-time adjustment, and efficient control of the hard shoulder.

[0004] Technical solution: To achieve the above-mentioned purpose, the present invention adopts the following technical solution:

[0005] In a first aspect, the present invention provides a method for dynamically opening a hard shoulder of a highway based on a NFGD model and a simulation platform, comprising the following steps:

[0006] Collect and use dynamic confidence mechanism to weightedly fuse multi-source real-time sensor data to obtain fused traffic state characteristics;

[0007] A dynamic control decision-making system for hard shoulder roads is based on the NFGD (Neuro-FLC, NSGA-II, DDPG) model, which includes a fuzzy-neural hybrid controller (Neuro-FLC), a multi-objective genetic algorithm controller (NSGA-II), and a reinforcement learning controller (DDPG). The system inputs a fusion state vector into the fuzzy-neural hybrid controller for real-time opening tendency judgment, outputs an opening tendency score, and determines whether to trigger strategy optimization. When triggering strategy optimization, a multi-objective genetic algorithm is used to collaboratively solve multiple conflicting objectives and output a set of Pareto optimal control strategies. The control parameters include opening length, vehicle type restriction, speed limit, and opening and closing period. A reinforcement learning controller is constructed using the perception state, the optimal control strategy set, and historical feedback. The controller outputs an action vector based on the deterministic policy gradient method, representing the currently recommended hard shoulder opening strategy combination. The optimal control strategy set serves as the action space basis of the reinforcement learning controller. The action command acts on the simulation system or the actual induction system. The system collects feedback indicators of the current cycle, iteratively optimizes the strategy network and value network, and forms a closed-loop adaptive control strategy.

[0008] Interact with the simulation platform in real time to implement control command issuance, feedback collection, and online strategy evaluation.

[0009] Furthermore, the fused traffic state features include average vehicle speed, queue length, compression, large truck occupancy rate, traffic density, and weather and accident risk information; observations of the same variable by multiple sensors By normalizing the weights w k (t) Weighted fusion; using dynamic confidence mechanism to calculate weights Normalized weights Satisfy the normalization constraint; where K represents the number of sensors, represents the observation value of the kth sensor on variable i, Var represents the variance of the observation value sequence in a certain time window, σ 2 Expressed as a normalization factor or reference variance constant, it is used to adjust the tolerance for fluctuations.

[0010] Furthermore, in the fuzzy-neural hybrid controller, the fuzzy inference layer has built-in expert rules and uses Gaussian membership functions to calculate the fuzzified values of input variables; the neural network optimization layer performs online training on the fuzzy rule parameters through a three-layer feedforward network, and the training data comes from the benefit comparison set of the opening strategy in the historical simulation; when the output opening tendency score is greater than the scoring threshold, or when the output opening tendency score is greater than the scoring threshold and the shoulder width and / or traffic risk meet the set conditions, the opening instruction is triggered and the control strategy is optimized.

[0011] Furthermore, in the collaborative solution of multiple conflicting objectives using a multi-objective genetic algorithm, the NSGA-II non-dominated sorting genetic algorithm is used to generate a multi-objective control strategy, and the optimization objectives include maximizing the average speed, minimizing the lane change disturbance, and minimizing the lane change disturbance; the algorithm uses mixed integer real number coding to represent the control parameters, and the chromosome structure includes the length of the open hard shoulder section, the type of vehicles allowed to pass, the corresponding section speed limit, the opening start time and the closing time; the algorithm first generates a Pareto optimal strategy set, and then selects the comprehensive optimal solution through the TOPSIS method as the reinforcement starting point of the reinforcement learning controller.

[0012] Furthermore, the state space of the reinforcement learning controller includes the average speed of each lane, queue length, traffic density, compression index, truck occupancy rate, and openness tendency score output by the fuzzy-neural hybrid controller. The action space constructs a discrete action set based on the Pareto solution set output by the genetic algorithm. The actions include the length of the open hard shoulder section, the type of vehicles allowed to pass, the corresponding section speed limit, the opening start time and closing time. The reward function takes into account the average vehicle speed, lane change disturbance intensity, and the accident risk factor estimated by simulation.

[0013] Furthermore, different expert rules are set and different models are trained for different scenarios such as the opening of the hard shoulder of the main line, the opening of the hard shoulder of the outgoing ramp, and the opening of the hard shoulder of the incoming ramp.

[0014] Furthermore, the simulation platform dynamically modifies simulation parameters by extending lane attribute labels and speed limit configurations, adopts multi-threading and dynamic step adjustment mechanisms to ensure synchronization between simulation time and actual time, and controls the simulated vehicle travel path through OD point setting.

[0015] In a second aspect, the present invention provides a highway hard shoulder dynamic opening system based on the NFGD model and simulation platform, which is used to implement the above-mentioned highway hard shoulder dynamic opening method based on the NFGD model and simulation platform, including:

[0016] The data transmission and acquisition module is used to collect and use a dynamic confidence mechanism to weightedly fuse multi-source real-time sensor data to obtain the fused traffic state characteristics;

[0017] The system control module is used to make dynamic control decisions for hard shoulder roads based on an NFGD model that includes a fuzzy-neural hybrid controller, a multi-objective genetic algorithm controller, and a reinforcement learning controller. The module includes: inputting a fusion state vector into the fuzzy-neural hybrid controller for real-time opening tendency judgment, outputting an opening tendency score, and determining whether to trigger strategy optimization; when triggering strategy optimization, using a multi-objective genetic algorithm to collaboratively solve multiple conflicting objectives and output a set of Pareto optimal control strategies, with control parameters including opening length, vehicle type restriction, speed limit, and opening and closing period; utilizing the perception state, the optimal control strategy set, and historical feedback to construct a reinforcement learning controller that outputs an action vector based on a deterministic policy gradient method, representing the currently recommended hard shoulder opening strategy combination, wherein the optimal control strategy set serves as the action space basis of the reinforcement learning controller; the action command acts on the simulation system or the actual induction system, and the system collects feedback indicators of the current cycle, iteratively optimizes the strategy network and value network, and forms a closed-loop adaptive control strategy;

[0018] The simulation module is used to interact with the simulation platform in real time to implement control command issuance, feedback collection, and online strategy evaluation and response.

[0019] In the third aspect, the present invention provides a computer system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the steps of the method for dynamic opening of hard shoulders of highways based on the NFGD model and simulation platform are implemented.

[0020] In a fourth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of a method for dynamically opening hard shoulders of highways based on the NFGD model and simulation platform.

[0021] Beneficial Effects: The proposed method and system for dynamic opening of hard shoulders on highways based on the NFGD model and simulation platform integrates multi-source perception, intelligent decision-making, and simulation linkage, and has the following technical advantages:

[0022] 1. Dynamic confidence mechanism improves fusion accuracy: By building a dynamic confidence weight model, the weighting coefficients are adaptively adjusted according to the historical stability and real-time credibility of various sensors, achieving highly robust fusion of traffic status characteristics and significantly enhancing perception accuracy.

[0023] 2. Multi-model collaboration improves decision-making intelligence: Build an NFGD model, integrate a fuzzy-neural hybrid controller, an NSGA-II multi-objective optimization controller, and a DDPG reinforcement learning controller to achieve real-time open tendency judgment, strategy generation, and closed-loop tuning, respectively, forming an intelligent control chain and effectively improving the adaptability and long-term stability of the control strategy.

[0024] 3. Real-time simulation linkage ensures timely response: The simulation platform enables online evaluation and real-time feedback of control strategies. Combined with multi-threading and dynamic step-size control mechanisms, this ensures synchronization between simulation time and actual time, improving the immediacy and verifiability of strategy responses. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flow chart of a dynamic opening method for hard shoulders of highways based on the NFGD model and simulation platform.

[0026] Figure 2 This is the simulation module flow chart.

[0027] Figure 3 This is the architecture diagram of the integrated sensing system.

[0028] Figure 4 Schematic diagram of the connection between network reliability and simulation.

[0029] Figure 5 This is a schematic diagram of the simulated road network of some sections of the Beijing-Shijiazhuang Expressway.

[0030] Figure 6 This is a flow chart of multi-threaded simulation operations.

[0031] Figure 7 Graphs showing the relationship between real time and simulation time in (a) single-threaded and (b) multi-threaded simulations.

[0032] Figure 8 Schematic diagram of the separation of simulated travel section units.

[0033] Figure 9 Schematic diagram of the hard shoulder control strategy in the integrated sensing system.

[0034] Figure 10 Example diagram of a simulation scenario for a two-way four-lane trunk road.

[0035] Figure 11 Example diagram of a simulation scenario for a two-way four-lane merging ramp.

[0036] Figure 12 Example of a simulation scenario for a two-way four-lane merging ramp. DETAILED DESCRIPTION

[0037] The technical solutions and technical effects of the present invention are further described below with reference to the accompanying drawings and specific embodiments.

[0038] An embodiment of the present invention discloses a method for dynamically opening the hard shoulder of a highway based on a NFGD model and a simulation platform, comprising: collecting and weightedly fusing multi-source real-time sensor data using a dynamic confidence mechanism to obtain fused traffic state characteristics; making dynamic control decisions for the hard shoulder based on the NFGD model including a fuzzy-neural hybrid controller, a multi-objective genetic algorithm controller, and a reinforcement learning controller; and interacting with the simulation platform in real time to implement control command issuance, feedback collection, and online strategy evaluation. Among them, the dynamic control decision-making process of the hard shoulder based on the NFGD model includes: inputting the fusion state vector into the fuzzy-neural hybrid controller for real-time opening tendency judgment, outputting the opening tendency score, and judging whether to trigger strategy optimization; when triggering strategy optimization, using a multi-objective genetic algorithm to collaboratively solve multiple conflicting goals, and outputting a set of Pareto optimal control strategy sets, the control parameters of which include opening length, vehicle type restriction, speed limit value and opening and closing period; using the perception state, optimal control strategy set and historical feedback to construct a reinforcement learning controller based on the deterministic policy gradient method to output an action vector, which represents the currently recommended hard shoulder opening strategy combination, where the optimal control strategy set serves as the action space basis of the reinforcement learning controller; the action command acts on the simulation system or the actual induction system, the system collects the current cycle feedback indicators, iteratively optimizes the strategy network and value network, and forms a closed-loop adaptive control strategy.

[0039] Accordingly, a dynamic open system for hard shoulder traffic on highways based on the NFGD model and simulation platform that implements the above method mainly includes the following core modules: a data transmission and acquisition module, which is used to collect and weightedly fuse multi-source real-time sensor data using a dynamic confidence mechanism to obtain the fused traffic state characteristics; a system control module, which is used to make dynamic control decisions for hard shoulder traffic based on the NFGD model including a fuzzy-neural hybrid controller, a multi-objective genetic algorithm controller and a reinforcement learning controller; and a simulation module, which is used to interact with the simulation platform in real time to realize the issuance of control instructions, feedback collection, and online strategy evaluation.

[0040] exist Figure 1 In the process, multi-source data is collected by sensors, and after transmission, communication analysis, system control and simulation verification, it is decided whether to execute the hard shoulder opening and generate the opening strategy.

[0041] exist Figure 2 In the simulation, the platform constructs a static road network and expands the policy tags. After achieving policy synchronization and time coordination, it connects to the Neuro-FLC and NSGA-II intelligent decision-making units. Finally, DDPG implements closed-loop tuning and TRACI executes the control strategy.

[0042] The following describes in detail the method and system for dynamically opening the hard shoulder of a highway based on the NFGD model and simulation platform according to an embodiment of the present invention.

[0043] 1. Architecture of road network sensing system

[0044] This embodiment takes into account the characteristics of the sensors connected to the system and the characteristics of the target being sensed, and determines the appropriate data transmission communication mode. The data processing process after data is transmitted to the system includes data preprocessing, data persistence, data mining, structured data storage and effective information output. According to the data processing flow, the sensor system module integrated with SUMO simulation is as follows: Figure 3 shown.

[0045] (1) Multi-source data transmission and acquisition module

[0046] It includes a hardware sensor group (coil detector, video detector, microwave radar) and a software sensor group (on-board OBU, web crawler, meteorological API), and builds a three-dimensional standardized classification framework based on temporal resolution, spatial distribution characteristics, and data structure;

[0047] (2) Communication and protocol module

[0048] Communication is a fundamental element of system construction, enabling the various components of the sensing system to work together. Further traffic management relies heavily on the collection of sensor information from various sensors. Data preprocessing is a key component of the communication and protocol module, including data decoding according to the protocol, data source verification, data quality verification, and the formation of a valid data structure.

[0049] (3) Database module

[0050] The database module provides a crucial function: persisting various types of data and enabling standard data queries. Data persistence is a crucial component of the system, storing data in an appropriate location and providing support for further data queries. The data that needs to be persisted falls into three categories: structured sensor data, configuration data, and static large content files.

[0051] (4) Client interaction module

[0052] The client interaction module facilitates interaction between the sensing system and the client. This includes processing underlying data based on business layer requirements, extracting valid information from raw data, and transmitting this information to the business layer. The client interaction module also translates client operations into actual system tasks, one of which is demand-based data mining. Depending on the purpose of data mining, data mining methods within the sensing system include real-time data analysis, scheduled data analysis, and historical data analysis. Real-time tasks involve immediate processing of real-time data, targeting vehicle motion status data and illegal snapshots. Scheduled data analysis analyzes traffic data collected cyclically over a certain period of time. Application scenarios include congestion analysis in specific areas and indicator calculation. Historical data analysis analyzes long-term historical data stored in the database, such as traffic demand distribution on weekdays and holidays. Data mining is implemented with the actual purpose of system construction in mind.

[0053] (5) Simulation module

[0054] The simulation module relies on the sensing system's actual sensor data, but also possesses independent features. The simulation module can operate the simulation process based on business needs, including generating simulation data and executing operations that control the simulation process, such as changes in traffic demand, signal timing, and vehicle insertion and removal. The inclusion of simulation makes the sensing system more flexible, efficient, highly visual, and easy to operate.

[0055] (6) System control module

[0056] The system control module, the core decision-making and control hub of this embodiment, is responsible for generating, triggering, and scheduling the hard shoulder opening control strategy. Based on fused traffic state feature inputs, this module invokes an algorithm controller to determine real-time opening trends. It then integrates with the strategy optimization module to generate multi-objective control solutions and implements closed-loop strategy iteration and updates through a reinforcement learning controller. The system control module also coordinates two-way communication with the simulation module to issue control commands, collect feedback metrics, and conduct real-time strategy evaluation, ensuring the timeliness, adaptability, and stability of the opening strategy.

[0057] (7) Interface display and operation module

[0058] The interface display and operation module provides a visual display of system operating status and access to manual intervention and control, supporting access from a variety of terminal devices. Its functions include dynamically displaying integrated traffic status indicators (such as average speed and queue length), open control strategy execution status, simulation feedback, and other core operational information. It also provides interactive features such as user authorization login, strategy parameter setting, and system mode switching (e.g., automatic, semi-automatic, or manual intervention). The module interface supports real-time graphical large-screen display, enabling management personnel to track and intervene in hard shoulder control situations within the highway operation monitoring center.

[0059] 2. Highway traffic state modeling and multi-source data fusion

[0060] (1) Data sources and types

[0061] The multi-source traffic data collection system involved in this embodiment includes data sources such as loop detectors, video detectors, microwave radars, OBU on-board units, web crawlers, API interfaces, and social platforms. The loop detectors collect point-based structured traffic flow, speed, and occupancy data at high frequency. The video detectors obtain semi-structured queue length, vehicle ID, and trajectory data at the high-frequency segment / region level. The microwave radar provides structured speed vectors and spatial distance data from high-frequency point-segment composite detection. The OBU on-board equipment records high-frequency regional-level structured GPS trajectory and lane-changing behavior data. The web crawlers collect low-frequency regional-level unstructured holiday forecast and emergency public opinion data. The API interfaces obtain low-frequency regional-level structured meteorological and weather emergency data. The social platform mines low-frequency regional-level unstructured congestion perception hot word data. The system uses three dimensions: temporal resolution (high frequency / low frequency), spatial distribution characteristics (point / segment / region), and data structure (structured / semi-structured / unstructured) to standardize and classify multi-source heterogeneous traffic data, establishing a unified metadata framework for data fusion processing.

[0062] Table 1 Multi-source data types

[0063] Data Source Collection data type Temporal resolution Spatial distribution characteristics Data Structure Coil detector Traffic flow, vehicle speed, and occupancy rate high frequency point Structural Video Detector Queue length, vehicle ID, trajectory high frequency Segment / Region Semi-structured microwave radar Velocity vector, spatial distance high frequency Point / Segment Structural OBU (Onboard Unit) GPS trajectory, lane-changing behavior high frequency area Structural Web crawlers Holiday forecasts and public opinion on emergencies low frequency area Unstructured API interface Meteorological data, weather emergencies (fog / rainfall) low frequency area Structural social platforms Congestion perception hot words low frequency area Unstructured

[0064] (2) Core modeling method

[0065] This embodiment extracts the following main state variables:

[0066] Average speed (per lane):

[0067]

[0068] in: is the average speed of the nth lane segment at time t (unit: km / h); is the total number of detected vehicles at time t on the nth lane segment; is the instantaneous speed of the jth vehicle on the lane segment at time t.

[0069] Queue length (per direction): calculated by image processing or historical traffic backlog model:

[0070]

[0071] in: is the queue length of the nth lane segment at time t (unit: m); is the length of the detection section corresponding to the nth lane segment (unit: m); is the traffic density of the nth lane segment at time t (unit: veh / km / lane); ρ cr is the critical density, which refers to the critical value at which the traffic state changes from free flow to congestion; ρ jam is the congestion density, which is the maximum density when the road is fully saturated.

[0072] Compression index:

[0073]

[0074] in: is the compression index of the nth lane segment, indicating the degree of traffic congestion; is the traffic density of the nth lane segment; is the average speed of the nth lane segment.

[0075] Large truck share:

[0076]

[0077] in: is the occupancy rate of large trucks in the mth area; N tru is the number of large trucks in the mth area; N total is the total number of vehicles in the mth area (including cars, trucks, etc.).

[0078] Traffic flow density:

[0079]

[0080] in: is the traffic density of the nth lane segment; N total is the total number of vehicles in the nth lane segment (including passenger cars, trucks, etc.); is the road section length corresponding to the nth lane segment (unit: m).

[0081] (3) Weighted fusion model

[0082] The observations of the same variable by multiple sensors are recorded as:

[0083]

[0084] in: is the fusion result of the i-th state variable; is the observation value of the k-th sensor on the i-th state variable; K is the total number of sensor categories participating in the observation of this variable; the fusion output is calculated as follows: Among them, w k(t) is the fusion weight of the k-th sensor at time t, satisfying: ∑w k (t)=1,w k (t)∈[0,1].

[0085] (4) Dynamic Confidence Mechanism

[0086] Each type of sensor is assigned a dynamic confidence level, which is calculated as follows:

[0087]

[0088] Where: k (t) is the dynamic confidence value of the k-th sensor at time t; The variable x is the k-th sensor pair i The variance of the observations within a certain time window; σ 2 The empirical standardized variance is used to adjust the sensitivity of confidence to fluctuations and can be set to the historical average variance.

[0089] The weights are related to the confidence normalization:

[0090]

[0091] (5) Output interface and format standardization

[0092] The final fusion state feature output is:

[0093]

[0094] Among them: n The nth monitored lane segment number is used for indexing variables such as speed, density, and compression; s m is the mth region number, used to count statistics such as the proportion of large trucks; z is the global number, used for macro-influencing factors such as weather and accidents; is the set of lane segments monitored at the current moment; The state vector is a set of all currently involved regions. The state vector is uniformly structured and encapsulated in JSON format and transmitted to the simulation and control platform through the TCP / UDP / WebSocket communication interface.

[0095] 3. Integration of actual road network sensing system and simulation

[0096] (1) Construction of simulation environment based on SUMO

[0097] Simulation is a method that simulates actual road networks and dynamic traffic flows. Compared to traditional mathematical traffic analysis methods, simulation offers a high degree of integration in vehicle following models, lane change models, and signal control models. Simulation comprehensively considers various factors of the road network and intuitively displays dynamic traffic conditions.

[0098] Considering the need for microscopic perception of traffic flow in the sensor system, this embodiment uses the SUMO simulation tool for road network simulation. SUMO has several important functions: controlling simulation speed, dynamically allocating traffic flow, and outputting various parameters.

[0099] (2) Synchronization of actual traffic status and simulation

[0100] The sensor system integrated with SUMO simulation can intuitively display the motion state of microscopic targets and fill in the missing parts of the road network. The key to integration is to synchronize the parameters of the actual road network with the simulation. The synchronization stage includes the synchronization of static road network structure, processing time, control strategy, sensor information and induction measures. The specific data flow and content of each stage are as follows: Figure 4 shown.

[0101] The static road network synchronization phase is a key component of building an integrated sensing system. This phase's primary function is to collect static road network data and complete static road network simulation modeling. Numerous map sources exist to describe the map structure, and SUMO simulation utilizes XML-formatted map data sources. Three map sources are provided based on actual needs: Open Street Map (OSM), a map editing GUI, and high-precision point data.

[0102] This embodiment uses XML-formatted map data to simulate the road network in SUMO simulation. The static information of the road network map includes the location of road sections, the location of weaving areas, road channelization, and the connection relationship between different elements.

[0103] Processing time synchronization is the dynamic basis for the integration of SUMO simulation and the actual road network sensor system. In the process of mapping the actual sensor state to the simulation, it is necessary to maintain the consistency between the actual time and the simulation time. Therefore, this embodiment proposes a method based on multi-threading and dynamic step adjustment to optimize the simulation processing time. The characteristic of SUMO simulation is that the simulation is divided into discrete processes based on the simulation step, and the simulation is advanced by controlling the simulation step. However, when the tasks performed in each step are different, different steps require different execution times, resulting in a large deviation between the actual simulation time and the actual time. The multi-threading and dynamic simulation step control method takes the simulation step execution content and the simulation step execution time as the control objects to achieve a rough correspondence between the simulation time and the actual time.

[0104] The control strategy synchronization phase involves mapping the actual control strategy to the simulation. This phase extracts valid information from the control strategy in the actual road network and implements a similar control strategy in the simulation. Synchronization ensures that the driving environment in the simulation resembles the actual road network. Highway network control strategies primarily include vehicle speed limits, road traffic sign settings, and vehicle type restrictions.

[0105] Sensor information synchronization is a further advancement in driving environment simulation. Traffic flow in a road network exhibits significant temporal and spatial distribution differences. The primary function of the sensor information synchronization step is to map the traffic flow distribution of the actual road network into the simulation system. Based on sensor characteristics, traffic flow information is converted into vehicle movement information in the simulation. Sensors are considered as traffic demand generators, and the locations of downstream sensors are considered as traffic demand destinations, creating point-to-point (OD) pairs in the simulation.

[0106] Synchronizing inductive measures is a further step in the synchronization of control strategies. This phase is performed after the simulation is complete. Its primary purpose is to verify the effectiveness of the control strategy by modifying it, such as by implementing hard shoulder opening and closing, vehicle type restrictions, and changes in traffic demand. By implementing inductive measures early in the simulation, comprehensive data can be obtained to support traffic management.

[0107] 4. Fuzzy-neural hybrid controller (Neuro-FLC) determines the hard shoulder opening conditions

[0108] In highway traffic guidance systems, determining in real time whether to open the hard shoulder to alleviate congestion is a core decision-making challenge. Traditional methods rely on static thresholds (such as when vehicle speeds fall below a certain threshold or when the proportion of large trucks exceeds a threshold), lack adaptability, and are difficult to adapt to complex dynamic scenarios. Furthermore, the default step-size control mechanism of simulation platforms like SUMO cannot guarantee high-precision synchronization between simulation time and real-time, affecting the timeliness of control strategies.

[0109] (1) Input variables

[0110] The Neuro-FLC module takes the state features after multi-source data fusion as input, namely average speed, queue length, compression index, large truck occupancy rate, traffic flow density, weather and accidents.

[0111] (2) Fuzzy logic reasoning

[0112] Example of setting rules for fuzzy controller:

[0113] IF Δv is High AND ΔQ is High THEN Openness = Strong

[0114] IF Δv is Medium AND ΔQ is Increasing THEN Openness = Moderate

[0115] The fuzzy output score is denoted as μ open ∈[0,1]

[0116] (3) Neural network self-learning

[0117] Neuro-FLC constructs a three-layer feedforward neural network (input layer, hidden layer, output layer) to train and update fuzzy membership function parameters. Training data comes from historical simulation cases where "opening strategies are better than not opening." It combines scores, road conditions, and risk indicators to output a dynamic threshold score. The final opening logic is:

[0118] Openifμ open >θ(t)∧W≥3.0m∧R risk <ε

[0119] Where: θ(t) is the dynamic scoring threshold; W is the hard shoulder width; R risk The traffic risk value may be predicted using any existing traffic risk prediction model, where ε is the risk threshold; in some other embodiments, the traffic risk value may not be considered.

[0120] 5. Multi-objective optimization strategy generation module (NSGA-II)

[0121] This example proposes an open hard shoulder control strategy using the NSGA-II (non-dominated sorting genetic algorithm) as its core optimization engine. This algorithm collaboratively solves multiple conflicting objectives and outputs a set of Pareto-optimal control strategies. This set of strategies serves as the action space foundation for the reinforcement learning controller, providing stable strategy candidates for the system's subsequent closed-loop self-learning.

[0122] This embodiment sets three types of key performance indicators as optimization targets

[0123] (1) Maximizing average traffic speed:

[0124]

[0125] Where: V avg is the average speed of all vehicles in the observation area (unit: km / h); is the number of all road sections in the observation area; is the average speed of the nth observation section.

[0126] (2) Minimizing congestion delay costs:

[0127]

[0128] Where: C delay is the congestion delay cost, which represents the time loss caused by the non-free flow state of vehicles in different time periods; T is the total number of time steps of the optimization evaluation (e.g., in minutes); is the average travel time of the mth region in the jth time step; is the benchmark travel time under free flow conditions in the area.

[0129] (3) Minimizing lane-changing disturbance intensity:

[0130]

[0131] Among them: F change is the average lane-changing disturbance intensity, which indicates the frequency of lane-changing behavior of a unit vehicle within a certain period of time; T is the total number of time steps corresponding to the observation duration; The total number of vehicles in this lane segment at time t; Is whether a lane change occurs in this lane segment at time t.

[0132] The final optimization objective function is defined as: max{V avg ,-C delay ,-F change}

[0133] To facilitate genetic algorithm processing, this embodiment encodes the control strategy parameters into a decision variable vector:

[0134] X=[L,C v ,V limit ,t open ,t close ]

[0135] Where: L is the length of the open hard shoulder section (unit: meter); C v The type of vehicles allowed to pass (enumeration value: 0-passenger car, 1-truck, 2-mixed traffic); V limit is the speed limit value of the corresponding section (unit: km / h); t open is the opening start time (seconds); t close is the closing time (seconds).

[0136] The above parameters are represented by mixed integer real number encoding to adapt to the NSGA-II chromosome structure.

[0137] (4) Definition of constraints

[0138] The width of the hard shoulder has a significant impact on driving behavior. The wider the hard shoulder, the faster the vehicle speed, and vice versa. See Table 2 for details.

[0139] Table 2 Effect of hard shoulder width on driving speed

[0140] Hard shoulder width Bus speed Truck speed 3.0m 100 / 3.5m 100 / 3.75m 120 80 4m 120 80 4.25m 120 80

[0141] This embodiment further analyzes the types of vehicles supported by hard shoulders of different widths, including three situations: no vehicles can pass, only passenger cars can pass, and passengers and freight can pass together. Vehicle speed limits are set according to different hard shoulder widths.

[0142] Table 3 Hard shoulder speed limits and vehicle restrictions

[0143] Design speed Hard shoulder width Passing vehicles Speed Limit (Document) Speed limit (simulation) 80km / h ＜3.0 No access / / 100-120km / h 3.0 Passenger car 80km / h 80km / h 120km / h 3.5 Passenger car 110 km / h 110km / h 120km / h 3.75 Mixed passenger and freight traffic 120 km / h 120km / h 120km / h 4.0 Mixed passenger and freight traffic 120 km / h 120km / h 120km / h 4.25 Mixed passenger and freight traffic 120 km / h 120km / h

[0144] To ensure the engineering feasibility of the generation strategy, this embodiment sets the following constraints:

[0145] The hard shoulder width must be ≥3.0m before it can be opened; the speed limit parameters must meet V limit ∈[80,120]; the opening and closing timing must meet t open ≤t close ; The open vehicle types should match the width conditions (such as only small passenger cars are allowed).

[0146] (5) Brief description of NSGA-II algorithm process

[0147] This example addresses the dynamic opening control problem of hard shoulders on highways by introducing the NSGA-II non-dominated sorting genetic algorithm into the control strategy generation module, combining it with traffic state perception and simulation evaluation mechanisms to achieve multi-objective optimization and control. The specific steps of the algorithm are as follows:

[0148] a. Individual definition and population initialization: According to the hard shoulder control requirements, each genetic individual is encoded as a five-dimensional vector, representing a set of candidate control parameter combinations: X i =[L,C v ,V limit ,t open ,t close ].

[0149] b. Objective function evaluation: Call the SUMO simulation platform to simulate the traffic operation results of each candidate solution and calculate three optimization indicators: {V avg ,-C delay ,-F change}, this process is based on the initial traffic state input provided by the perception system to ensure that the control strategy is adapted to the current road conditions.

[0150] c. Non-dominated sorting and crowding distance calculation: Sort the population based on the Pareto dominance relationship to form several frontier levels, while maintaining population diversity through the crowding distance indicator.

[0151] d. Evolutionary operations and new population generation: We use an elite retention strategy to select outstanding individuals, generate a new generation of individuals through tournament selection, simulated binary crossover (SBX), and polynomial mutation, and continuously optimize the quality of the solution.

[0152] e. Convergence judgment and solution set output: When the maximum number of iterations is reached or the convergence threshold is met, the non-dominated frontier set in the current population is output as the final Pareto optimal strategy set, covering typical control objective combinations such as "high traffic efficiency, low lane change disturbance, and low congestion delay."

[0153] f. In this embodiment, the NSGA-II genetic algorithm is used to generate a set of Pareto-optimal control policies based on real-time traffic state assessments during each control cycle. These policies include a combination of parameters such as opening length, vehicle type restrictions, speed limits, and opening and closing times. This set of policies is not directly used as controller output, but rather as a candidate set for initializing the action space of the reinforcement learning controller.

[0154] 6. Closed-loop feedback and reinforcement learning strategy iteration mechanism

[0155] To further enhance the intelligence and adaptability of the highway hard shoulder open control system, this embodiment proposes a closed-loop strategy iteration mechanism based on reinforcement learning, based on multi-source data feature fusion and multi-objective strategy optimization. This mechanism dynamically adjusts the induction strategy through continuous interaction, and has the ability to improve long-term performance.

[0156] Specifically, the system uses the traffic state feature vector obtained by the current fusion perception as the input of the reinforcement learning controller, including the average speed of each lane, queue length, traffic density, compression index, truck occupancy rate, and the open propensity score output by the fuzzy controller.

[0157] The reinforcement learning controller (such as DDPG) uses the optimal solution set output by the NSGA-II genetic algorithm to construct its optional action set A, namely: Among them: A t is the action candidate space at time t; is the kth Pareto optimal strategy output by NSGA-II; s t is the current environment state (provided by the perception fusion module).

[0158] On this basis, the reinforcement learning controller outputs an action vector based on the deterministic policy gradient method (DDPG), which represents the currently recommended hard shoulder opening strategy combination: a t ={L,C v ,V limit ,t open ,t close}.

[0159] The action command acts on the simulation system or the actual induction system through the TRACI interface. The system collects the feedback indicators of the current cycle, including the average vehicle speed V avg , lane-changing disturbance intensity F change , and the accident risk factor R estimated by simulationaccident . From this, we construct the immediate reward function: r t =ω1V avg -ω2F chang -ω3R accident ; Among them, ω1, ω2, ω3 are target weighting coefficients, which support flexible adjustment according to traffic goals.

[0160] The controller transforms the current state, action, reward and next state into a quadruple (s t ,a t ,r t ,s t+1 ) is stored in an experience pool, and parameters are updated via a policy network (Actor) and a value network (Critic). Both are constructed using a two-layer fully connected neural network. The Actor network takes traffic status as input and outputs continuous control actions; the Critic network takes a combination of status and action as input and outputs a corresponding Q-value estimate. The value network minimizes mean squared loss based on the target Q-value, while the policy network maximizes the Q-value output using policy gradients. To improve learning stability, the system introduces a soft update target network and combines it with a temporal difference method for reinforcement training.

[0161] The reinforcement learning controller proposed in this embodiment is highly compatible with the SUMO simulation system, completing a policy evaluation and update within each simulation control cycle (30 to 120 seconds), supporting both online learning and offline batch training. The system also uses the NSGA-II optimal policy as the initial DDPG action space, improving convergence efficiency and policy interpretability.

[0162] The following is an illustrative example of the solution and its effects using a real-world scenario simulation.

[0163] This example uses a section of the Beijing-Shijiazhuang Expressway as an example. The section features 32 pairs of surveillance cameras, 32 radio broadcast systems, 20 microwave vehicle detectors, and other hardware and software detectors. The loop detectors detect parameters such as traffic flow, queue length, and time intervals. The video detectors on this section are high-pixel cameras that use video frame processing algorithms to capture specific traffic parameters. This allows for the acquisition of target information, such as speed, unique identifiers, and time intervals within a specific area. This provides real-time sensor data for SUMO simulations.

[0164] 1. System composition and architecture

[0165] Multi-source data transmission and acquisition module: includes hardware sensors (coil and radar detectors, video detectors) and software sensors, used to collect parameters such as traffic flow, vehicle speed, lane occupancy, etc.

[0166] Communication and protocol module: Use long connections (such as WEBSOCKET) to achieve active data push, combined with HTTP query mode for data interaction in low real-time scenarios; use UDP communication for data with higher real-time requirements (coil data, radar data, etc.), and use TCP or MQTT protocols for scenarios with higher data integrity requirements.

[0167] Database module: used to persist various sensor data, configuration data and large content files, and provide support for subsequent data mining.

[0168] Client interaction module: converts raw sensor data into effective information that meets business needs, and realizes intelligent judgment of traffic status through data mining (real-time, timed and historical analysis).

[0169] Simulation Module: Built on the SUMO simulation platform, it simulates traffic conditions on real-world road networks. It uses XML-formatted map data to implement static road network modeling, and utilizes multithreading and dynamic step-size control to synchronize simulation time with real-world time. The car-following model uses the Krauss modified model, and the lane-changing model is LC2013.

[0170] 2. Multi-source sensor data fusion and feature extraction

[0171] (1) Standardization of heterogeneous data

[0172] Hardware sensors: The coil detector uploads structured traffic, vehicle speed, and occupancy data at a frequency of 1 Hz and transmits it to the edge computing node in real time via the UDP protocol. The video detector transmits semi-structured video frame data via RTSP streaming, extracts vehicle ID, trajectory, and queue length, and outputs JSON format data packets.

[0173] Mobile terminal: The OBU device uploads GPS trajectory and lane change behavior data at a frequency of 0.5Hz through the 5GC-V2X communication module, using Protobuf encoding for compressed transmission.

[0174] Internet data: Web crawlers collect hot congestion keywords from platforms such as Weibo and Amap every hour, and extract semantic features using the BERT-NLP model; the meteorological API obtains regional weather data every 15 minutes and stores it in a structured manner in a MySQL database.

[0175] (2) Dynamic weight fusion calculation

[0176] Based on the above weight distribution model, a dynamic weight distribution matrix is constructed. When the microwave radar detects sudden rainfall, its confidence weight α radar The weight of the video detector is increased to 0.4, while the weight of the video detector is reduced to 0.2 to cope with image recognition errors caused by reduced visibility.

[0177] The fusion output state vector is pushed to the SUMO simulation platform through the WebSocket interface. The data format example is as follows:

[0178] {"timestamp":"2024-03-20T14:30:00Z","speed_avg":65.2,"queue_length":320,"density":28.5,"weather":"rain","incident":false,"shoulder_score":0.78}.

[0179] 3. Integration and synchronization process of sensor data and simulation platform

[0180] The system includes the following stages in the data synchronization process:

[0181] (1) Static road network synchronization phase: The static data of the actual road network in the target area (main road sections, weaving areas, etc.) is collected through OpenStreetMap, and the XML format map data that meets the requirements of SUMO is generated. The map elements are then modified using the map editing GUI.

[0182] (2) Processing time synchronization stage: Dynamic processing time synchronization is the basis for subsequent operations after static road network synchronization. To ensure that the simulation processing time is consistent with the actual time, the multi-threaded operation mode in JAVA is used to control the simulation time. The main thread in the multi-threaded operation is used to realize the simulation step advancement and time synchronization, and any other operations required by the simulation business are performed through the secondary thread. The simulation step interval of the main thread is controlled by the thread sleep time operation according to the simulation speed, so that the processing time between the simulation and the actual is basically consistent. The multi-threaded operation process is implemented as follows Figure 6 As shown in the figure. The multi-threaded operation mode in the simulation system can help the sensor system control the simulation progress. After the simulation task division and task multi-threading operation, the simulation time in the sensor system changes as shown in the figure. Figure 7 As shown in the figure, it can be seen that multi-threaded operation in simulation can reduce the simulation time loss caused by user operation and make the simulation time basically consistent with the actual time.

[0183] (3) Control strategy synchronization phase: Speed limit control and traffic strategy control are performed during control strategy synchronization. By limiting the speed of specific road sections and specific vehicle types, the maximum speed limit control is the same as the actual road network. The TRACI interface provided by SUMO is used to map the speed limits, road signs, vehicle type restrictions and other control strategies in the actual road network into the simulation environment to ensure consistency between the simulation environment and the actual road network. The speed limit in the area is 120 km / h for passenger cars and 80 km / h for trucks. During the hard shoulder control synchronization process, UDP technology is used to access the actual variable information board to publish the control strategy.

[0184] (4) In the sensor information synchronization stage, the weaving area sensor data and the road section sensor data are analyzed and converted into corresponding motion state vehicles. These simulated vehicles are put into the simulation for synchronous movement. The sensing area is regarded as a node to separate the target moving area in the simulation. The upstream and downstream sensors are regarded as the generation point and destination of traffic demand, and OD points are created in the simulation to intuitively display the traffic status of the road network and divide the spatial simulation units into Figure 8 shown.

[0185] The simulated vehicle travel uses the sensing area as the feature to separate the spatial travel segments. The segment camera detector collects the vehicle information of the segment, and the collected vehicle information provides the speed and unique identification information for the simulated vehicle. Therefore, the segment of the upstream detector is set as the travel demand generation point, the downstream exit detector is used as the travel demand destination, and the segment between the generation point and the destination is used as the simulation unit of the simulated vehicle. Figure 8 As shown, the simulated vehicle completes a simulated unit of travel in the simulated road section 1.

[0186] (5) Inducement measure synchronization stage: After completing the above synchronization stage, the process of building a simulation module from the actual road network is completed. Inducement measure synchronization is to map the actual inducing measures to the simulation. Improving the quality of travel services and traffic management structure is an important goal of building a sensing system. The integration of simulation and actual road network sensing systems can more comprehensively and intuitively display the actual road conditions. In the integrated sensing system, the inducing measure of temporarily opening the hard shoulder is carried out, such as Figure 9 As shown in the figure, real-time and historical traffic data collected by each sensor is analyzed to determine the opening thresholds for main roads, merging, and exiting scenarios. After the simulation is complete, a dynamic hard shoulder control strategy is fed back based on real-time data to verify its effectiveness.

[0187] 4. Intelligent decision-making and control strategy generation

[0188] (1) Neuro-FLC dynamic decision-making

[0189] The input speed fluctuation rate and compression index are Z-score normalized to eliminate dimensional differences. 32 expert rules are built in, such as:

[0190] IF vehicle speed fluctuation rate IS High (μ=0.7) AND queue length change rate IS Rising (μ=0.6)

[0191] THEN Openness = Strong (output score + 0.5)

[0192] A training set (input: traffic state vector, output: optimal opening decision) is constructed based on historical data, and the Adam optimizer is used to iteratively update the membership function parameters.

[0193] (2) NSGA-II multi-objective optimization

[0194] The control strategy parameters use mixed encoding. Chromosome = [350,0,80,3600,7200] means that a 350-meter hard shoulder is open, allowing only small passenger cars to pass, with a speed limit of 80 km / h, and the open period is from 3600 seconds to 7200 seconds. A penalty function is applied to individuals that violate the hard shoulder width-vehicle type matching:

[0195] if width<3.75m and vehicle_type! =0:

[0196] fitness*=0.5#penalty function weakens fitness

[0197] Pareto frontier screening: After 50 iterations, 15 non-dominated solution sets are output, and the optimal overall strategy is selected using the TOPSIS method. In the early stages of system deployment, or before the reinforcement learning controller has completed policy network training, the system uses the TOPSIS method to calculate the closeness index of each solution from the multiple Pareto solutions output by NSGA-II, based on the weighted priorities set by the management (such as efficiency or safety). The strategy with the highest closeness is selected as the current control instruction.

[0198] The selection process is as follows:

[0199] a. Construct a decision matrix: All Pareto non-dominated solutions are calculated according to the three target indicators {V avg ,-C delay ,-F change}Construct a standardized decision matrix, where the negative sign indicates that the cost item needs to be minimized.

[0200] b. Normalization processing: Perform range normalization processing on each target indicator to form a dimensionless evaluation matrix.

[0201] c. Set target weights: Set the weights of each target based on the preferences of the road management department, for example, ω = [0.4, 0.3, 0.3], corresponding to traffic efficiency, delay cost, and lane change disturbance, respectively.

[0202] d. Construct ideal and negative ideal solutions: define the theoretical optimal solution (i.e. the optimal value of each target indicator) and the worst solution. Calculate the Euclidean distance between each solution and the ideal solution. And the Euclidean distance between each solution and the negative ideal solution Calculate relative closeness: Where: T i The larger it is, the closer the solution is to the optimal solution.

[0203] e. Select the optimal strategy: select the one with the maximum closeness T i The individual is used as the comprehensive optimal control strategy for the current cycle and as the reinforcement starting point of the DDPG controller.

[0204] (3) Reinforcement learning closed-loop tuning

[0205] As the system's operating cycles accumulate, a reinforcement learning controller (using the DDPG algorithm) constructs a state-action-reward triple based on traffic perception, the NSGA-II output policy set, and feedback metrics. This continuous interaction trains the policy network (Actor) and the value network (Critic). During this phase, the controller no longer relies on static TOPSIS selection, but instead adaptively selects actions from the Pareto solution set.

[0206] The input vector of the state space contains six features: [speed_avg, queue_length, density, truck_ratio, shoulder_score, weather_level]. The action space is constructed based on the Pareto solution set output by NSGA-II to construct a discrete action set, where action = [open length, vehicle type restriction, speed limit, opening time, closing time] ∈ {200m ≤ length ≤ 500m, 0 / 1 / 2, 60 / 80 / 100km / h, 1800-3600s, 3601-6000s}. The reward function is a weighted combination of three indicators:

[0207] reward=0.6×Δspeed-0.3×lane_change_penalty

[0208] -0.1×risk_factor

[0209] The experience pool uses prioritized experience replay and has a storage capacity of 10^5 transfer samples. The actor-critic network uses a dual-objective network structure and synchronizes parameters every 100 steps to prevent overfitting.

[0210] 5. Dynamic hard shoulder control execution

[0211] (1) Instruction issuance

[0212] Push the optimal strategy to the roadside unit (RSU) via the MQTT protocol to drive the variable message sign (CMS) and lane control signs: {Topic: / highway / shoulder_control

[0213] Payload:{"action":"open","length":350,"speed_limit":80,"valid_time":3600}}.

[0214] (2) Real-time control algorithm

[0215] For the dynamic opening control of the hard shoulder, the system has designed the following control process:

[0216] Step 1: Expand in SUMO network file <lane>Tag attributes, set the hard shoulder to only allow specific types of vehicles to pass, and define dynamic speed limit parameters;

[0217] <lane id="shoulder_1"width="3.0"allow="passenger" / >

[0218] <vClassSpeed dev="0.2"vClass="passenger"speed="33" / >

[0219] <vClassSpeed dev="0.2"vClass="truck"speed="22" / >

[0220] Step 2: Obtain real-time sensor data through the main thread loop. According to the control strategy, once the conditions are met, use the TRACI interface to issue control instructions and modify the simulation parameters in real time.

[0221] Step 3: At the same time, a dynamic simulation step control algorithm is used to keep the simulation time synchronized with the actual time, thereby achieving real-time dynamic control.

[0222] The specific algorithm pseudo code example is as follows:

[0223] Table 4 Dynamic simulation Python pseudo code

[0224]

[0225] Experimental analysis and system advantages

[0226] In order to verify the universal applicability of this embodiment, three typical scenarios for opening hard shoulders in expressway networks were determined, namely, opening the hard shoulder of the main line, opening the hard shoulder of the outgoing ramp, and opening the hard shoulder of the incoming ramp.

[0227] Opening the hard shoulder on the main line: This is primarily intended to expand free-flow scenarios, reduce interference between vehicles within the traffic flow, and allow vehicles to travel closer to the speed limit. This primarily addresses the issue of large trucks occupying multiple lanes, which reduces overall driving efficiency.

[0228] Experimental results of this embodiment in two-way 4-lane and 6-lane highway scenarios show that the intelligent control system using Neuro-FLC dynamic decision-making, NSGA-II multi-objective optimization, and DDPG reinforcement learning significantly improves road traffic efficiency. In the two-way 4-lane scenario, the system dynamically generates the optimal opening strategy (passenger car speed limit of 100km / h) by real-time evaluation of vehicle speed fluctuation and queue change rate, which increases the average speed from 76.6km / h to 83.6km / h (an increase of 9.13%), while reducing lane change disturbances by 31.2%, and keeping the accident risk index below the safety threshold of 0.15. For the two-way 6-lane scenario, the system adopts a mixed passenger and freight traffic strategy (truck speed limit of 80km / h), which increases the average speed from 78.7km / h to 84.8km / h (an increase of 7.75%), and reduces lane change disturbances by 28.5%. This effectively alleviates road congestion and improves traffic efficiency.

[0229] Table 5 Results of opening the hard shoulder of the main line

[0230]

[0231] Opening the hard shoulder of the outgoing ramp: This is mainly used in scenarios where outgoing vehicles are queuing and affecting normal driving on the main road. A certain distance upstream of the outgoing ramp will prompt that only vehicles exiting the ramp can enter the open hard shoulder, preventing straight-moving vehicles from entering.

[0232] In a two-way, four-lane merging scenario, the system dynamically generates the optimal opening strategy (80 km / h speed limit for passenger cars) by real-time evaluating ramp demand ratios and mainline traffic density. This strategy increased the average speed in the merging area from 58.3 km / h to 65.7 km / h (a 12.7% increase), reduced merging conflicts by 37.6%, and maintained the accident risk index below the safety threshold of 0.18. For a two-way, six-lane scenario, the system adopts a time-sharing mixed traffic strategy for passenger and freight traffic (passenger cars during peak hours), increasing the average speed in the merging area from 61.2 km / h to 67.5 km / h (a 10.3% increase), and reducing merging disturbances by 32.8%. This effectively alleviated mainline traffic pressure and ensured vehicle traffic efficiency and road section operation stability.

[0233] Table 6 Export of hard shoulder opening results

[0234]

[0235]

[0236] Opening the hard shoulder on the merging ramp: This is primarily intended for scenarios where merging vehicles intersect and disrupt normal traffic on the main road. Traffic on the hard shoulder upstream of the merging point is prohibited to avoid excessive conflict with ramp traffic. Downstream of the merging point, the hard shoulder is opened for speeding and merging, with a specific length.

[0237] In a two-way, four-lane merging scenario, the system dynamically generates an optimal lane opening strategy (prioritizing passenger cars) by real-time assessing mainline traffic flow (2450 pcu / h) and ramp merging demand (800 pcu / h). This strategy increases the average speed on the hard shoulder section to 73.4 km / h, bringing the mainline capacity to 3050 pcu / h (a 24.4% increase), while reducing merging conflicts by 39.2% and keeping the accident risk index below 0.17. For a two-way, six-lane scenario, the system uses a dynamic lane allocation strategy based on a mainline flow of 3750 pcu / h and a merging flow of 800 pcu / h. This strategy maintains an average speed of 73.6 km / h, increases capacity to 4460 pcu / h (an 18.8% increase), and reduces merging disturbances by 35.6%. This process allows right-turning vehicles to quickly merge onto the main road using the hard shoulder, effectively alleviating congestion and improving traffic efficiency.

[0238] Table 7 Results of hard shoulder opening

[0239]

[0240]

[0241] An embodiment of the present invention also discloses a computer system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the computer system implements any one of the aforementioned steps of the method for dynamically opening the hard shoulder of a highway based on the NFGD model and simulation platform.

[0242] An embodiment of the present invention also discloses a computer program product, including a computer program, which, when executed by a processor, implements the steps of any of the aforementioned methods for dynamically opening hard shoulders of highways based on the NFGD model and simulation platform.

[0243] The above embodiments are only preferred implementations of the present invention. Those skilled in the art may make various modifications and variations to the above embodiments without departing from the spirit and scope of protection of the present invention, and all of these modifications and variations should be considered to be within the scope of protection of the present invention.< / lane>

Claims

1. A dynamic opening method for hard shoulders of highways based on NFGD model and simulation platform, characterized by: The steps include: Collect and use dynamic confidence mechanism to weightedly fuse multi-source real-time sensor data to obtain fused traffic state characteristics; Based on the NFGD model including a fuzzy-neural hybrid controller, a multi-objective genetic algorithm controller and a reinforcement learning controller, dynamic control decision-making of the hard shoulder is carried out, including: inputting the fusion state vector into the fuzzy-neural hybrid controller for real-time opening tendency judgment, outputting the opening tendency score, and judging whether to trigger strategy optimization; when triggering strategy optimization, using a multi-objective genetic algorithm to collaboratively solve multiple conflicting objectives and output a set of Pareto optimal control strategy sets, with control parameters including opening length, vehicle type restriction, speed limit value and opening and closing period; using the perception state, the optimal control strategy set and historical feedback to construct a reinforcement learning controller based on the deterministic policy gradient method to output an action vector, which represents the currently recommended hard shoulder opening strategy combination, where the optimal control strategy set serves as the action space basis of the reinforcement learning controller; the action command acts on the simulation system or the actual induction system, the system collects the feedback indicators of the current cycle, iteratively optimizes the strategy network and value network, and forms a closed-loop adaptive control strategy; Interact with the simulation platform in real time to implement control command issuance, feedback collection, and online strategy evaluation.

2. The method for dynamic opening of hard shoulders of highways based on NFGD model and simulation platform according to claim 1 is characterized in that: The fused traffic state characteristics include average vehicle speed, queue length, compression, large truck occupancy rate, traffic density, and weather and accident risk information; Multiple sensors observing the same variable By normalizing the weights w k (t) Weighted fusion; using dynamic confidence mechanism to calculate weights Normalized weights Satisfy the normalization constraint; where K represents the number of sensors, represents the observation value of the kth sensor on variable i, Var represents the variance of the observation value sequence in a certain time window, σ 2 Expressed as a normalization factor or reference variance constant, it is used to adjust the tolerance for fluctuations.

3. The method for dynamic opening of hard shoulders of highways based on NFGD model and simulation platform according to claim 1 is characterized in that: In the fuzzy-neural hybrid controller, the fuzzy inference layer has built-in expert rules and uses Gaussian membership functions to calculate the fuzzified values of input variables. The neural network optimization layer uses a three-layer feedforward network to perform online training on fuzzy rule parameters, with training data coming from a benefit comparison set of opening strategies in historical simulations. When the output opening propensity score is greater than a scoring threshold, or when the output opening propensity score is greater than the scoring threshold and the shoulder width and / or traffic risk meet set conditions, an opening instruction is triggered to optimize the control strategy.

4. The method for dynamic opening of hard shoulders of highways based on NFGD model and simulation platform according to claim 1 is characterized in that: In the collaborative solution of multiple conflicting objectives using a multi-objective genetic algorithm, a NSGA-II non-dominated sorting genetic algorithm is used to generate a multi-objective control strategy, and the optimization objectives include maximizing average speed, minimizing lane change disturbance, and minimizing lane change disturbance; The algorithm uses mixed integer real number coding to represent control parameters. The chromosome structure includes the length of the open hard shoulder section, the type of vehicles allowed to pass, the corresponding section speed limit, the opening start time and the closing time. The algorithm first generates a Pareto optimal strategy set and then selects the comprehensive optimal solution through the TOPSIS method as the reinforcement starting point of the reinforcement learning controller.

5. The method for dynamic opening of hard shoulders of highways based on NFGD model and simulation platform according to claim 1 is characterized in that: The state space of the reinforcement learning controller includes the average speed of each lane, queue length, traffic density, compression index, truck occupancy rate, and the openness propensity score output by the fuzzy-neural hybrid controller. The action space constructs a discrete action set based on the Pareto solution set output by the genetic algorithm. The actions include the length of the open hard shoulder section, the type of vehicles allowed to pass, the corresponding section speed limit, the opening start time and closing time. The reward function takes into account the average vehicle speed, the lane change disturbance intensity, and the accident risk factor estimated by simulation.

6. The method for dynamic opening of hard shoulders of highways based on NFGD model and simulation platform according to claim 1 is characterized in that: Different expert rules are set and different models are trained for different scenarios such as the opening of the hard shoulder of the main line, the opening of the hard shoulder of the outgoing ramp, and the opening of the hard shoulder of the incoming ramp.

7. The method for dynamic opening of hard shoulders of highways based on NFGD model and simulation platform according to claim 1 is characterized in that: The simulation platform dynamically modifies simulation parameters by extending lane attribute labels and speed limit configurations, adopts multi-threading and dynamic step adjustment mechanisms to ensure the synchronization of simulation time with actual time, and controls the simulated vehicle travel path through OD point setting.

8. A highway hard shoulder dynamic opening system based on the NFGD model and simulation platform, used to implement the highway hard shoulder dynamic opening method based on the NFGD model and simulation platform according to any one of claims 1 to 7, characterized in that: include: The data transmission and acquisition module is used to collect and use a dynamic confidence mechanism to weightedly fuse multi-source real-time sensor data to obtain the fused traffic state characteristics; The system control module is used to make dynamic control decisions for hard shoulder roads based on an NFGD model that includes a fuzzy-neural hybrid controller, a multi-objective genetic algorithm controller, and a reinforcement learning controller. The module includes: inputting a fusion state vector into the fuzzy-neural hybrid controller for real-time opening tendency judgment, outputting an opening tendency score, and determining whether to trigger strategy optimization; when triggering strategy optimization, using a multi-objective genetic algorithm to collaboratively solve multiple conflicting objectives and output a set of Pareto optimal control strategies, with control parameters including opening length, vehicle type restriction, speed limit, and opening and closing period; utilizing the perception state, the optimal control strategy set, and historical feedback to construct a reinforcement learning controller that outputs an action vector based on a deterministic policy gradient method, representing the currently recommended hard shoulder opening strategy combination, wherein the optimal control strategy set serves as the action space basis of the reinforcement learning controller; the action command acts on the simulation system or the actual induction system, and the system collects feedback indicators of the current cycle, iteratively optimizes the strategy network and value network, and forms a closed-loop adaptive control strategy; The simulation module is used to interact with the simulation platform in real time to implement control instruction issuance, feedback collection, and online strategy evaluation.

9. A computer system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the computer program is executed by a processor, the steps of a method for dynamically opening a hard shoulder of a highway based on an NFGD model and a simulation platform are implemented according to any one of claims 1 to 7.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of a method for dynamically opening a hard shoulder of a highway based on an NFGD model and a simulation platform are implemented according to any one of claims 1 to 7.

Images