Main line adaptive coordination control method based on multi-source data fusion
By using multi-source data fusion and neural network prediction to predict vehicle arrival time, and dynamically optimizing the cycle and green wave ratio, the problem of insufficient dynamic adaptability of traffic flow in existing trunk line coordinated control methods is solved, thereby improving the utilization rate of green wave bandwidth and traffic efficiency.
Patent Information
- Application Number
- CN202511728924.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-24
- Publication Date
- 2026-02-06
AI Technical Summary
Existing arterial coordination control methods are difficult to adapt to the dynamic fluctuations of traffic flow. Fixed cycles lead to lagging or inaccurate control strategies. The utilization rate of green wave bandwidth is low. The balance of traffic demand at each approach lane at the intersection is ignored. Existing control models fail to effectively integrate spatiotemporal correlation characteristics and lack accurate prediction of vehicle arrival times.
By employing a multi-source data fusion approach, real-time traffic data is acquired through electronic police and radar-visual integrated machines. Combined with long short-term memory neural networks and graph convolutional neural networks, vehicle arrival times are predicted, control periods are dynamically divided, the cycle and green light ratio are optimized, an adaptive coordinated control scheme is generated, and the phase difference and green light duration are adjusted in real time.
It enables accurate prediction and dynamic adjustment of traffic flow, improves the utilization rate of green wave bandwidth and trunk line traffic efficiency, is compatible with existing signal systems, requires no large-scale modification, and is suitable for complex traffic scenarios.
Smart Images
Figure CN121483039A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of urban intelligent transportation technology, specifically to a trunk line adaptive coordinated control method based on multi-source data fusion. Background Technology
[0002] Traditional arterial coordination methods, such as MAXBAND and PASSER, are mostly based on fixed time periods and historical average traffic flow to design timing schemes, making it difficult to adapt to dynamic fluctuations in traffic flow. Although existing adaptive coordination systems introduce coils or geomagnetic detectors to obtain real-time data, they are limited by issues such as limited detector coverage, high installation and maintenance costs, and susceptibility to data quality interference, resulting in lagging or inaccurate control strategies.
[0003] Current mainstream adaptive systems generally use fixed common cycle constraints. Once the cycle is set, each intersection can only adjust the green light duration within that cycle, and cannot dynamically adjust the cycle itself based on real-time traffic flow. If the cycle is forcibly changed, the original phase difference relationship is disrupted, and green wave coordination fails. In addition, phase differences are mostly calculated based on preset vehicle speeds, without considering the spatiotemporal changes in actual vehicle arrival times, resulting in low green wave bandwidth utilization. At the same time, green light ratio optimization often ignores the balance of traffic demand at each approach lane of the intersection, causing excessive delays in some directions.
[0004] Existing technologies, such as integrated radar-visual systems, can acquire real-time parameters like vehicle location, speed, and queue length with high precision. Electronic police and checkpoint systems have also accumulated massive amounts of historical traffic data, providing a foundation for multi-source fusion. However, existing control models have not effectively integrated spatiotemporal correlation features and lack the ability to accurately predict vehicle arrival times; control periods still rely on manual experience for division, making them unable to respond to sudden changes in traffic conditions; and optimization objectives often focus on single indicators, neglecting the coordination between green light time utilization efficiency and intersection balance. Summary of the Invention
[0005] (a) Technical problems to be solved
[0006] To address the shortcomings of existing technologies, this invention provides a trunk line adaptive coordinated control method based on multi-source data fusion.
[0007] (II) Technical Solution
[0008] To achieve the above objectives, the present invention provides the following technical solution: A trunk line adaptive coordinated control method based on multi-source data fusion, comprising the following steps:
[0009] Step 1: Multi-source traffic data collection and preprocessing: Collect real-time traffic flow data, clean and denoise the data, remove outliers, and construct a standardized traffic data set;
[0010] Step 2, Traffic Network Graph Construction and Feature Extraction: Based on the road network topology, construct a traffic network graph;
[0011] Extract the spatiotemporal features of nodes, including historical traffic flow, weather data, traffic time periods, real-time traffic flow parameters, road attributes, and traffic light status;
[0012] Step 3, Vehicle arrival time prediction: A fusion model of long short-term memory neural network and graph convolutional neural network is adopted;
[0013] Step 4: Control period division and constraint parameter determination: Based on traffic flow parameters from multi-source traffic data, hierarchical clustering is used to divide the control period.
[0014] Step 5: Background scheme generation: Without changing the phase sequence of the original signal timing, calculate the optimal cycle and green ratio of each intersection in each control period based on historical traffic flow data;
[0015] Step 6, Dynamic Phase Difference Adjustment: Collect the distance between upstream and downstream intersections on the main road and the vehicle arrival time predicted in Step 3, and calculate the dynamic phase difference between adjacent intersections.
[0016] Step 7, Adaptive Optimization of Cycle and Green Light Ratio: Based on real-time traffic flow data detected by the integrated radar-visual system, an optimization model is established with the goal of maximizing green light utilization. The optimal cycle C and the green light duration for each phase are then calculated. ;
[0017] Step 8, Signal control scheme generation and execution: Integrate the dynamically adjusted phase difference, optimized cycle and green ratio to generate an adaptive coordinated control scheme for the trunk line, and send it to the signal controllers at each intersection for execution; collect traffic flow feedback data in real time and execute steps 3-7 in a loop.
[0018] Preferably, the multi-source traffic data includes historical traffic flow data from electronic police and checkpoint equipment, as well as real-time traffic flow data collected by the radar-visual integrated machine, including vehicle location, speed, direction of movement, and queue length.
[0019] More preferably, the nodes of the traffic network diagram represent road segments or intersections, the edges represent road connections, and the edge weights are assigned to at least one of traffic flow, vehicle speed, road length, road curvature, and road slope.
[0020] Preferably, the long short-term memory neural network mines time series correlations, and the graph convolutional neural network processes spatial dependencies.
[0021] Preferably, the constraint parameters include the maximum green light time for each phase. Minimum green light time and adjustable time intervals [ , ],in Let be the initial green light duration for the i-th phase in the background scheme.
[0022] More preferably, the objective function for optimization in step 7 is: ,in: C is the period. denoted as the green light duration for each phase, and j as the number of phases in the signal scheme.
[0023] Preferably, the optimized constraints include phase sequence constraints, maximum green light time constraints, minimum green light time constraints, equalization constraints for each approach lane at the intersection, and adjustable phase time constraints.
[0024] (III) Beneficial Effects
[0025] Compared with existing technologies, this invention provides a trunk line adaptive coordination control method based on multi-source data fusion, which has the following beneficial effects:
[0026] This technical solution integrates multi-dimensional information such as vehicle location, speed, and queue length collected in real time by electronic police, historical data from checkpoints, and radar-visual integrated machines to construct a high-quality standardized traffic dataset. It also combines road network topology to establish a weighted traffic network graph, comprehensively depicting the spatiotemporal dynamic characteristics.
[0027] An LSTM-GCN fusion model is used to accurately predict vehicle arrival times, taking into account both time series evolution patterns and spatial road network dependencies, reducing prediction errors by more than 15%. Hierarchical clustering is used to dynamically divide control time periods, avoiding the disconnect between fixed-time-period strategies and actual traffic demand. While preserving the original phase sequence, a background timing scheme is generated, and a dynamic phase difference calculation mechanism is introduced to achieve precise matching of traffic flow and signals.
[0028] With the goal of maximizing green light utilization, a system considering G is established. max / G min An optimization model with multiple constraints, including adjustable phase range and equalization of approach lanes, is used to solve for the optimal cycle and green light ratio in real time. The control scheme is updated every 30–60 seconds in closed-loop iteration, significantly improving the traffic efficiency of trunk lines. This method is compatible with existing signal systems, requires no large-scale modification, and is suitable for complex traffic scenarios such as urban arterial roads and expressways, providing an efficient, adaptive, and scalable coordinated control solution for intelligent transportation. Attached Figure Description
[0029] Figure 1 This is a schematic diagram of the overall process architecture of the present invention;
[0030] Figure 2 This is a schematic diagram of the traffic network architecture of the present invention. Detailed Implementation
[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0032] Please see Figure 1-2 The present invention provides a trunk line adaptive coordinated control method based on multi-source data fusion, comprising the following steps:
[0033] Step 1: Multi-source traffic data collection and preprocessing: Collect real-time traffic flow data, clean and denoise the data, remove outliers, and construct a standardized traffic data set;
[0034] Step 2, Traffic Network Graph Construction and Feature Extraction: Based on the road network topology, construct a traffic network graph;
[0035] Extract the spatiotemporal features of nodes, including historical traffic flow, weather data, traffic time periods, real-time traffic flow parameters, road attributes, and traffic light status;
[0036] Step 3, Vehicle arrival time prediction: A fusion model of long short-term memory neural network and graph convolutional neural network is adopted;
[0037] Step 4: Control period division and constraint parameter determination: Based on traffic flow parameters from multi-source traffic data, hierarchical clustering is used to divide the control period.
[0038] Step 5: Background scheme generation: Without changing the phase sequence of the original signal timing, calculate the optimal cycle and green ratio of each intersection in each control period based on historical traffic flow data;
[0039] Step 6, Dynamic Phase Difference Adjustment: Collect the distance between upstream and downstream intersections on the main road and the vehicle arrival time predicted in Step 3, and calculate the dynamic phase difference between adjacent intersections.
[0040] Step 7, Adaptive Optimization of Cycle and Green Light Ratio: Based on real-time traffic flow data detected by the integrated radar-visual system, an optimization model is established with the goal of maximizing green light utilization. The optimal cycle C and the green light duration for each phase are then calculated. ;
[0041] Step 8, Signal control scheme generation and execution: Integrate the dynamically adjusted phase difference, optimized cycle and green ratio to generate an adaptive coordinated control scheme for the trunk line, and send it to the signal controllers at each intersection for execution; collect traffic flow feedback data in real time and execute steps 3-7 in a loop.
[0042] This technical solution is based on the core logic of multi-source data fusion, accurate prediction, dynamic constraints, full-parameter optimization, and closed-loop iteration. By integrating historical traffic data with real-time data from the radar-visual integrated machine, a spatiotemporal correlation prediction model is constructed to obtain vehicle arrival times. Combined with hierarchical clustering to divide control periods and clarify constraint boundaries, it achieves full-parameter adaptive optimization of cycle, green light ratio, and phase difference without changing the phase sequence. It generates a trunk line coordinated control scheme and continuously iterates and adjusts it. The core advantage lies in breaking through the limitations of rigid parameter adjustment in traditional control. Through data-driven dynamic optimization, it maximizes green wave bandwidth and green light utilization, improves trunk line traffic efficiency, and solves the technical pain point of coordination failure after changes in the common cycle.
[0043] Multi-source traffic data acquisition and preprocessing solution
[0044] Multi-source traffic data includes historical traffic flow data from electronic police and checkpoint equipment, as well as real-time traffic flow data such as vehicle location, speed, direction of movement, and queue length collected by the radar-visual integrated machine; the data is cleaned, denoised, and outlier removed to construct a standardized set.
[0045] Historical data provides support for long-term traffic flow patterns, reflecting traffic characteristics under different time periods and weather conditions, laying the foundation for the division of control time periods and the generation of background plans; the real-time data from the integrated radar-visual machine has the characteristics of high dimensionality and high precision, and can dynamically capture instantaneous changes in traffic flow, such as sudden congestion and increased queue length, providing a basis for real-time parameter optimization.
[0046] Data preprocessing uses methods such as mean imputation and the 3σ criterion to eliminate the interference of missing and outlier values on model prediction and optimization, ensuring data quality and guaranteeing the accuracy of subsequent steps.
[0047] Transportation Network Map Construction Scheme
[0048] In a traffic network diagram, nodes represent road segments or intersections, and edges represent road connections. The edge weight is assigned as at least one of traffic flow, vehicle speed, road length, road curvature, and road gradient.
[0049] A network graph is constructed based on the road network topology to intuitively present the connection relationship and spatial distribution of trunk roads. The purpose of assigning multi-dimensional values to edge weights is to comprehensively reflect the difficulty of road passage and traffic conditions. For example, road passages with high traffic volume and steep slopes have higher weights, which can be considered in the prediction of vehicle arrival time, making the prediction results more consistent with the actual road network conditions.
[0050] Vehicle arrival time prediction model
[0051] A fusion model of Long Short-Term Memory Neural Network (LSTM) and Graph Convolutional Neural Network (GCN) is adopted. LSTM mines time series correlations, while GCN handles spatial dependencies.
[0052] LSTM excels at capturing long-distance dependencies in the temporal dimension, analyzing time-series patterns such as morning peak congestion delays and smooth traffic flow during off-peak hours in historical traffic flow data, and predicting the temporal evolution trend of traffic flow. GCN focuses on spatial correlation mining, identifying spatial dependencies such as the transmission of congestion from upstream intersections to downstream sections and traffic diversion on parallel road segments in the road network. Combined with the weight information of the traffic network graph, it optimizes the prediction of vehicle arrival times between different nodes. The fusion of the two achieves spatiotemporal dual-dimensional prediction, avoiding the limitations of a single model that only considers time or spatial factors, keeping the vehicle arrival time prediction error within 5%, and providing accurate input for dynamic adjustment of phase difference.
[0053] Control period division and constraint parameter determination scheme
[0054] Hierarchical clustering is used to divide control time periods, with constraint parameters including the maximum green light time for each phase. Minimum green light time and adjustable time intervals [ , ], Let represent the initial green light duration for the i-th phase in the background scheme.
[0055] Hierarchical clustering classifies time periods with similar traffic flow characteristics into one category based on parameters such as flow rate and speed from multi-source traffic data, such as morning peak, off-peak, and evening peak, so that the traffic status within the same time period is relatively stable and frequent adjustments to control parameters are avoided.
[0056] and The phase is determined based on factors such as road capacity and pedestrian crossing needs, such as the phase of a main road. =60s, =20s, ensuring that the green light duration meets traffic demand without wasting resources; the adjustable time interval clarifies the adjustment boundary of the green light duration for each phase, ensuring the flexibility of parameter optimization while avoiding traffic chaos caused by excessive adjustment, and providing a constraint basis for subsequent cycle and green light ratio optimization.
[0057] Cycle and Green Ratio Optimization Model
[0058] With the goal of maximizing green light utilization, the objective function is: ,in: C is the period. Let j be the green light duration for each phase, and j be the number of phases in the signal scheme. The constraints include phase sequence constraints, maximum / minimum green light time constraints, equalization constraints for each approach lane at the intersection, and adjustable phase time constraints.
[0059] In the objective function: This represents the average green light utilization rate for each phase, reflecting the overall green light utilization efficiency. The standard deviation of utilization rate reflects the balance of green light allocation in each phase; the combination of the two achieves the dual goals of overall efficiency and local balance, avoiding both green light waste and congestion caused by insufficient green light duration in a certain phase.
[0060] Constraints ensure the rationality of optimization: Phase sequence constraints maintain the original traffic order and avoid conflicts caused by changes in phase sequence; maximum / minimum green light time constraints and adjustable time constraints limit the adjustment range; and equalization constraints for each approach lane at the intersection ensure that traffic flows in different directions can obtain reasonable green light durations, avoiding one-way congestion.
[0061] Detailed Workflow
[0062] Step 1: Multi-source traffic data collection and preprocessing
[0063] Historical traffic flow data from electronic police and checkpoint equipment is collected through the public security video network and stored in the historical traffic flow database, such as data from the past three months, as well as real-time data from the integrated radar and video cameras at various intersections on main roads, such as vehicle position, speed, direction of movement, and queue length, with a sampling frequency of 10Hz.
[0064] Missing data were processed using the mean imputation method, and outliers exceeding the normal range, such as data with instantaneous speed of 0 or far exceeding the road speed limit, were removed using the 3σ criterion to construct a standardized traffic data set.
[0065] Step 2: Traffic Network Graph Construction and Feature Extraction
[0066] Based on the trunk road network topology, each intersection and connecting road segment is set as a node, the road connection relationship is set as an edge, and the edge weight is assigned as traffic flow × road length to comprehensively reflect the traffic pressure of the road segment.
[0067] Extract the spatiotemporal features of nodes, including historical traffic flow, average flow for the corresponding time period, weather data such as sunny / rainy / snowy, traffic time period such as morning peak / off-peak, real-time traffic flow parameters such as current queue length and average speed, road attributes such as length, curvature, and slope, and traffic light status such as current phase and remaining green light time.
[0068] Step 3: Vehicle Arrival Time Prediction
[0069] Build an LSTM-GCN fusion model. Set two hidden layers in the LSTM layer, such as 64 and 32 neurons, and set one hidden layer in the GCN layer, such as 32 neurons. Train the model with historical data, iterate 100 times, and set a learning rate of 0.001.
[0070] The extracted spatiotemporal features are input into the trained model to predict the arrival time of vehicles at intersections upstream and downstream of the trunk line, and the prediction results are output.
[0071] Step 4: Control period division and constraint parameter determination
[0072] Based on parameters such as traffic flow and speed from multi-source traffic data, hierarchical clustering is used to divide the day into 5 control periods, such as morning peak 7:00-9:00, off-peak 9:00-11:30, noon peak 11:30-13:30, evening peak 17:00-19:00, and nighttime 19:00-7:00 the next day.
[0073] Taking into account road capacity and pedestrian crossing demand, calculate the traffic capacity of each phase in each time period. and
[0074] such as the morning rush hour phase of main roads =60s =20s, based on the initial green light duration Pi of the subsequent background scheme, determine the adjustable time interval for each phase. , ].
[0075] Step 5: Background scheme generation
[0076] Without changing the phase sequence of the original signal timing, the optimal cycle of each intersection in each control period is calculated based on historical traffic flow data, such as 120s for intersection A and 110s for intersection B, and the green light ratio.
[0077] Select the maximum value of the optimal cycle for all intersections on the trunk green wave coordinated road, such as 120s, as the common cycle, and adjust the duration of each phase at each intersection to ensure that it does not exceed the subsequently determined adjustable range, thus generating the background scheme for trunk adaptive coordinated control.
[0078] Step 6: Dynamic adjustment of phase difference
[0079] Collect the actual distance between the upstream and downstream intersections on the main road, such as 800m at intersection AB and 1000m at intersection BC, and compare it with the vehicle arrival time predicted in step 3.
[0080] Based on the logic that the time for a vehicle to arrive at the downstream intersection = the time for the upstream green light to turn on + the phase difference, the dynamic phase difference between adjacent intersections is calculated. If the predicted arrival time of the vehicle is 10 seconds earlier, the phase difference is shortened by 10 seconds to ensure that the vehicle arrives just in time to meet the green light, thus maintaining the green wave coordination effect.
[0081] Step 7: Adaptive Optimization of Period and Green Ratio
[0082] Collect real-time traffic flow data from the integrated radar-visual machine. For example, if the queue length of a certain phase increases by 30% compared to the initial state, substitute it into the optimization objective function.
[0083] Under the following constraints—namely, maintaining the phase sequence, ensuring the green light duration is within [Gmin, Gmax], conforming to the adjustable range, and maintaining balance among all entrance lanes—solve for the optimal period C and the green light duration for each phase.
[0084] For example, the cycle time can be adjusted from 120s to 130s, and the green light duration for congested phases can be adjusted from 40s to 50s.
[0085] Step 8: Signal control scheme generation and execution
[0086] By integrating the dynamically adjusted phase difference, optimized cycle, and green ratio, the final trunk adaptive coordinated control scheme is generated and distributed to the signal controllers at each intersection via the network for execution.
[0087] Real-time traffic flow feedback data, such as traffic delays, number of stops, and queue length, is collected. Steps 3-7 are executed in a loop every 5 minutes. Control parameters are continuously optimized based on the feedback data to achieve closed-loop adaptive adjustment.
[0088] Although embodiments of the invention have been shown, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A trunk line adaptive coordinated control method based on multi-source data fusion, characterized in that, Includes the following steps: Step 1: Multi-source traffic data collection and preprocessing: Collect real-time traffic flow data, clean and denoise the data, remove outliers, and construct a standardized traffic data set; Step 2, Traffic Network Graph Construction and Feature Extraction: Based on the road network topology, construct a traffic network graph; Extract the spatiotemporal features of nodes, including historical traffic flow, weather data, traffic time periods, real-time traffic flow parameters, road attributes, and traffic light status; Step 3, Vehicle arrival time prediction: A fusion model of long short-term memory neural network and graph convolutional neural network is adopted; Step 4: Control period division and constraint parameter determination: Based on traffic flow parameters from multi-source traffic data, hierarchical clustering is used to divide the control period. Step 5: Background scheme generation: Without changing the phase sequence of the original signal timing, calculate the optimal cycle and green ratio of each intersection in each control period based on historical traffic flow data; Step 6, Dynamic Phase Difference Adjustment: Collect the distance between upstream and downstream intersections on the main road and the vehicle arrival time predicted in Step 3, and calculate the dynamic phase difference between adjacent intersections. Step 7, Adaptive Optimization of Cycle and Green Light Ratio: Based on real-time traffic flow data detected by the integrated radar-visual system, an optimization model is established with the goal of maximizing green light utilization. The optimal cycle C and the green light duration for each phase are then calculated. ; Step 8, Signal control scheme generation and execution: Integrate the dynamically adjusted phase difference, optimized cycle and green ratio to generate an adaptive coordinated control scheme for the trunk line, and send it to the signal controllers at each intersection for execution; collect traffic flow feedback data in real time and execute steps 3-7 in a loop.
2. The trunk line adaptive coordinated control method based on multi-source data fusion according to claim 1, characterized in that, The multi-source traffic data includes historical traffic flow data from electronic police and checkpoint equipment, as well as real-time traffic flow data collected by the radar-visual integrated machine, including vehicle location, speed, direction of movement, and queue length.
3. The trunk line adaptive coordinated control method based on multi-source data fusion according to claim 1, characterized in that, The nodes of the traffic network diagram represent road segments or intersections, and the edges represent road connections. The edge weights are assigned to at least one of the following: traffic flow, vehicle speed, road length, road curvature, and road slope.
4. The trunk line adaptive coordinated control method based on multi-source data fusion according to claim 1, characterized in that, The long short-term memory neural network mines time series correlations, and the graph convolutional neural network processes spatial dependencies.
5. The trunk line adaptive coordinated control method based on multi-source data fusion according to claim 1, characterized in that, The constraint parameters include the maximum green light time for each phase. Minimum green light time and adjustable time intervals [ , ],in Let be the initial green light duration for the i-th phase in the background scheme.
6. The trunk line adaptive coordinated control method based on multi-source data fusion according to claim 1, characterized in that, The objective function for optimization in step 7 is: ,in: C is the period. denoted as the green light duration for each phase, and j as the number of phases in the signal scheme.
7. The trunk line adaptive coordinated control method based on multi-source data fusion according to claim 6, characterized in that, The optimized constraints include phase sequence constraints, maximum green light time constraints, minimum green light time constraints, equalization constraints for each approach lane at the intersection, and adjustable phase time constraints.