Intelligent traffic regulation and control method and system for city

Abstract

The invention discloses an intelligent traffic regulation and control method and system for cities, relates to the technical field of intelligent traffic, and realizes high-precision real-time sensing of complex traffic states of urban intersections by constructing a multi-source traffic data acquisition system. A multi-modal feature fusion and semantic tag matching mechanism is adopted to recognize overlapping and dynamic changes of traffic states, a multi-strategy linkage signal timing scheme is generated in time, and the problems that a traditional signal control system is lagged in response and single in state recognition are solved. According to the method, commuting and logistics interference indexes are introduced, a release sequence adjustment and channel optimization strategy is intelligently generated according to a people flow peak and logistics concurrence situation, and the method aims at solving the problem of intersection signal timing regulation and control under the condition of complex multi-mode traffic flow superposition. And meanwhile, an execution result is collected in a closed-loop manner, a strategy deviation value is calculated in real time, error tolerance is set, adaptive iterative optimization is supported, and stable convergence of the strategy and achievement of a regulation target are ensured.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, specifically to a smart traffic control method and system for cities. Background Technology

[0002] As urban transportation networks become increasingly complex, the deep integration of public transportation systems, multimodal travel modes (such as buses, subways, and shared bicycles), and urban logistics systems has become a typical trend. Particularly at subway entrances and exits, bus transfer hubs, or concentrated commercial areas, complex traffic conditions have emerged with a large number of concurrent and overlapping flows of people, vehicles, and goods. This condition is characterized by high spatiotemporal coupling and nonlinear fluctuations, and traditional traffic control methods have many limitations in addressing it.

[0003] On the one hand, most existing traffic signal control strategies are set based on single-dimensional traffic data, making it difficult to model the dynamic relationships between people, vehicles, and non-motorized vehicles. When peak pedestrian flow coincides with non-motorized vehicle return flow (such as the morning and evening tidal travel of shared bicycles) or when logistics vehicles suddenly enter the main road section, traditional systems have difficulty in timely identifying the "complexity" of the situation, which can easily lead to delayed signal configuration response, resulting in increased traffic congestion at intersections and conflicts between pedestrians and vehicles.

[0004] On the other hand, most traffic management platforms currently lack support for multi-state overlap identification and multi-strategy linkage control mechanisms. Especially in scenarios where logistics vehicle flow and commuter / non-motorized vehicle flow are simultaneously enhanced, the lack of interference index assessment and response models leads to logistics lanes being "occupied without yielding to pedestrians" for extended periods, or "forced insertion of logistics convoys" during peak commuting hours, exacerbating local congestion and even causing safety hazards. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a smart traffic control method and system for cities, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a smart traffic control method for cities, comprising the following steps: Step 1: Multi-source traffic data acquisition and adaptive preprocessing: Deploy Bluetooth detectors, video cameras, non-motorized vehicle detection loops, and logistics lane flow meters at key intersections in multi-modal traffic convergence areas to collect real-time data on Bluetooth passenger numbers N(t), single vehicle entry and exit volume S(t), and logistics vehicle flow Q(t) and their geographical location information; adaptively adjust the sampling frequency based on the detected network bandwidth and signal stability to generate a multi-modal dataset; Step 2: Establish and distribute standard traffic state semantics and template library: Based on the bus-subway-shared bicycle transfer scenario, construct a structured set of traffic states L and a corresponding signal timing template library T; distribute L and T to the controllers of each intersection. Step 3: Real-time Multimodal Fusion Recognition and Traffic State Detection: Normalize and time-align the multimodal dataset, project it onto a unified feature dimension to obtain the fused feature vector F(τ); perform similarity matching between F(τ) and the standard semantic label L and template library T to construct a multimodal overlap index. And assess, if >0.6, generate the first adjustment strategy; Step 4: Generate functional traffic control instructions and perform closed-loop iterative optimization: Construct a logistics and commuting interference index. The system then evaluates and generates second and third adjustment strategies, while simultaneously collecting real-time data on the improvement in intersection traffic efficiency. The deviation value ‖E‖ from the current strategy is set, and an error tolerance Eth is preset. If the deviation ‖E‖ exceeds Eth, parameter self-learning is triggered, L and T are updated, and iterative optimization is achieved until ‖E‖ converges to the allowable range.

[0007] Preferably, the multi-source traffic data collection in step one includes the following sub-steps: S11. Install Bluetooth detectors, video cameras, non-motorized vehicle detection coils, and logistics vehicle flow monitors at key intersections in multi-modal traffic areas to collect data on Bluetooth users N(t), single vehicle entry and exit S(t), and logistics vehicle flow Q(t). S12, and configure network time synchronization modules in edge nodes, including roadside computing devices, to align the timestamps of data from various sensing devices through the master-slave clock mechanism PTP protocol.

[0008] Preferably, the adaptive preprocessing in step one includes the following sub-steps: S13. Periodically detect and record the data distortion status indicators of each channel, including Bluetooth recognition jitter, video image blur, logistics vehicle detection packet loss rate and bandwidth utilization; integrate the quality indicators of each modal sensing signal in a weighted manner to construct a multimodal data comprehensive distortion index S; S14. A first threshold Sth1 and a second lower threshold Sth2 are preset. If the multimodal data comprehensive distortion index S exceeds the first threshold Sth1, it indicates that the current data transmission command is unqualified, and there is a risk of network congestion or sensor data redundancy. The first throttling sampling strategy is then executed, including: The image frame rate of the video image channel is reduced by 25% to 50%, from 30fps to 20fps or 15fps; the polling cycle of the Bluetooth identification channel is extended by 30% to 60%; the data upload interval of the non-motorized vehicle detector / logistics detector is extended by 20% to 40%, and an edge caching mechanism is enabled, uploading only when the detection result changes exceed the threshold, in order to compress the total data traffic; when S exceeds Sth1, the following transmission order is followed, including: video channel compression priority > Bluetooth scanning priority > detector reporting priority; S15. If the multimodal data comprehensive distortion index S is lower than the second lower limit threshold Sth2, it indicates that the current data transmission command is qualified and resources are abundant, and the conditions for executing high-frequency, high-fidelity sampling strategies are met. Therefore, the second sampling strategy is triggered, including: restoring the image frame rate of the video image channel to a proportional 30fps, and increasing the image resolution from 720p to 1080p; prioritizing an increase of 10% in the video sampling rate of mixed pedestrian and vehicle areas; shortening the polling cycle of the Bluetooth identification channel by 15% to 30%; restoring the data upload frequency of non-motorized vehicle detectors / logistics detectors to the normal frequency, and enabling a short-cycle fluctuation value early warning mechanism, especially when the morning and evening peak hours are approaching, increasing the reporting frequency by about 10% to 25%; if the shared bicycle entry and exit volume S(t) changes drastically by ±40%, the sampling frequency priority of the non-motorized vehicle lane channel will be temporarily increased, while other channels will remain unchanged. If the multimodal data overall distortion index S falls within the range of the first threshold Sth1 and the second lower threshold Sth2, it indicates that the current data transmission instruction is qualified and no adjustment is required. S16. After the first throttling sampling strategy or the second sampling strategy is executed, a multimodal dataset is generated.

[0009] Preferably, the specific steps for constructing the structured set L of traffic states in step two are as follows: S21. Decompose the typical traffic scenarios in the bus-subway-shared bicycle transfer process, divide the flow of people, vehicles, signal control, and logistics lane occupation into a "state-trigger-response" triplet, and collect at least 30 days of historical multimodal traffic data, including Bluetooth number of people N(t), bicycle entry and exit volume S(t), and logistics vehicle flow Q(t). S22. Perform sliding window processing on each type of traffic data, specifically including Bluetooth passenger count N(t), single vehicle entry / exit count S(t), and logistics vehicle flow Q(t). Set the time window to 5 minutes and the sliding step size to 1 minute. Extract each time segment as a state unit, with the expression as follows: ; This represents the k-th state segment. This represents the traffic data collected at time t, including any one or more vectors among Bluetooth users N(t), single vehicle entry and exit S(t), and logistics vehicle flow Q(t); Indicates the current window The time range starts from the beginning time. Start, duration is All data falling within this time frame belongs to this state segment; S23. For each intersection in each time segment, calculate the first derivative of the number of pedestrians via Bluetooth at time t: ; express The absolute value of the change in the number of people over a period of time; divided by This is to calculate the average rate of change, in units of people per minute; S24. For each intersection in each time segment, calculate the second derivative of the number of pedestrians via Bluetooth at time t: ; First, calculate the difference of the first derivatives of the two consecutive time windows, i.e., the change in the growth rate, and then divide by... Acceleration is obtained; S25. Calculate the total flow of shared bicycles S(t) at time t as: S(t) = Sin(t) + Sout(t); Sin(t) represents the number of shared bicycles entering at time t, and Sout(t) represents the number of shared bicycles leaving at time t. S26. Calculate the rate of change of logistics vehicle traffic within the time window using the first derivative of logistics vehicle traffic: ; This represents the flow of logistics vehicles collected at time t.

[0010] Preferably, step two, in constructing the structured set L of traffic states, further includes: S27, Preset first threshold Second threshold and the third threshold If, within three consecutive time windows, i.e., the current time slice τ∈{t,t−Δt,t−2Δt}, the following conditions are met: The first derivative of the number of pedestrians via Bluetooth satisfies ; The second derivative of the number of pedestrians via Bluetooth satisfies Generate a "high-density transfer" status label for this time period [t−2Δt, t]. If the total flow of shared bicycles S(t) is relative to the maximum capacity Cbike of the road segment, then: ; When the duration of this condition exceeds the minimum duration threshold of peak single-vehicle conditions, a "single-vehicle tidal return" status label is generated. If the first derivative of the logistics vehicle flow rate satisfies: ; If the corresponding vehicles are mainly concentrated at adjacent intersections less than 2km apart, a "static intersection congestion" status label will be generated. S29. Organize the semantic labels of the status labels within each time interval that meets the conditions into a structured set L of traffic status. , to The semantic label represents the first to the nth semantic label that meets the conditions; the structured semantic label set L contains multiple semantic labels, each label is accompanied by a set of standard state feature vectors generated by summarizing historical multimodal traffic data, which are used to characterize the standard state features of the semantic label in multiple dimensions such as Bluetooth user changes, single vehicle tidal degree, logistics vehicle fluctuation and mixed traffic degree. Each semantic tag includes: tag type, time interval, corresponding intersection code or geographic location information, and feature variable values ​​supporting tag judgment; each semantic tag in the traffic state structured set L matches one or more signal timing templates. This establishes a mapping relationship and constructs a corresponding signal timing template library T. , to This represents the signal timing template from the first signal timing template to the nth signal timing template that meets the conditions for each semantic tag; Signal timing template This includes parameters such as green light duration, red light delay, phase switching sequence, and priority parameters for specific lanes.

[0011] Preferably, the specific steps for deploying and calling the template library in step two are as follows: The constructed traffic state structure set L and the corresponding signal timing template library T are encapsulated into a unified data structure to generate the first control index table, specifically in SON or XML format, which facilitates network transmission and embedded decoding; the data is pushed to the central control platform via an edge router using serial bus RS485 or CAN, network protocols IP / TCP, MQTT, or WebSocket. The centralized control platform reads the first control index table and extracts the k-th state segment of each sampling window. If the tag condition is met, and tag L is matched, immediately invoke the bound signal timing template. ,include: Change the green light duration, start the left turn phase in advance, give priority to non-motorized vehicles, and coordinate the start of traffic lights.

[0012] Preferably, the specific steps of real-time multimodal overlap fusion recognition in step three are as follows: S31. A multimodal dataset obtained from edge nodes or sensing devices, including Bluetooth user count N(t), shared bicycle entry and exit volume S(t), video recognition pedestrian density V(t), and logistics vehicle flow Q(t), is executed as follows: Time alignment processing ensures that data from each channel corresponds to the same state segment at the same time t. Feature normalization processing normalizes various types of data to a unified interval [0,1]. Projecting onto a unified feature space generates a fused feature vector: S32. Perform cosine similarity matching between each fused feature vector and the standard state features in the traffic state structure set L to find the matching state label, and call the corresponding matching signal timing template T. i ; If multiple tags simultaneously meet the matching conditions, they are recorded as a multi-tag state set. Real-time collection and fusion of multi-source traffic data to form the current time slice Feature vector Calculate cosine similarity : ; If there exists an 's'-th element that satisfies the condition, then label it. satisfy If the value is ≥0.85, it is considered a "successful match", and then the matching signal timing template T is called accordingly. i .

[0013] Preferably, the specific steps of real-time multimodal overlap fusion recognition in step three also include: S33. If the same intersection is in the current time slice When matching multiple state labels, the weight corresponding to the j-th semantic label is used. ∈[0,1], calculate the multi-mode overlap index : ; in, This represents the total number of semantic tags that were successfully matched. Indicates the first The weight of each semantic label; If the multi-mode overlap index A value greater than 0.6 indicates multiple overlapping states, triggering the first adjustment strategy. This strategy prioritizes the use of a "multi-mode collaborative template," extending the green light duration for pedestrian lights, non-motorized vehicle lights, and freight lights by 10%–15%, 8%–12%, and 5%, respectively. It also advances the left-turn phase by 5–7 seconds to reduce intersection waiting conflicts. Non-motorized vehicles are given priority as first-level right-of-way, ensuring their passage during peak pedestrian and bicycle traffic periods. Furthermore, a coordination strategy with adjacent intersection traffic lights is implemented to improve tidal flow traffic by at least 18%. If the multi-mode overlap index A value of <0.3 indicates that the current traffic condition is clearly defined and has a single characteristic, and the previous control period should be maintained without needing to be advanced.

[0014] Preferably, step four includes: S41. If the same intersection is in the current time slice Within the area, there is a simultaneous increase in logistics vehicle traffic and pedestrian / non-motorized vehicle traffic; therefore, the logistics and commuting interference index is calculated. : ; in, This indicates the capacity of the mixed flow channel, which is preset based on the intersection's geographical structure, number of lanes, and width of pedestrian and non-motorized vehicle lanes. , and Each is a current time slice Bluetooth user count, bicycle entry and exit volume, and logistics vehicle traffic; S42, if A value of ≥1.2 indicates that the flow of logistics vehicles has posed a first risk of interference to commuter / non-motorized vehicle traffic, generating a second adjustment strategy, including: delaying the release phase of logistics vehicles, reducing the release time per cycle by 20% to 30%; designating 30% of the original freight lanes as "time-limited slow-moving lanes" or "temporary U-turn lanes" for a duration of no less than 15 minutes; increasing the green light duration by 12% to 20%; and responding to pedestrian crossing signals 3 seconds earlier. If the interference continues for more than 3 cycles, with each cycle lasting 5 minutes, then a structural traffic restriction mode will be implemented, requiring logistics vehicles at the intersection to detour via the auxiliary road. If 0.8≤ ≤1.2 indicates a second risk of interference between pedestrian and vehicle traffic, generating a third adjustment strategy, including: adjusting the release delay of logistics vehicles by 10% to 15%; extending the green light switching buffer by 3 to 5 seconds to prevent overlapping conflicts between pedestrians and vehicles; releasing shared bicycles 5 seconds earlier with the green light; and triggering pop-up guidance to avoid logistics channels in densely populated shared bicycle sections. like A value of <0.8 indicates that there is no risk of mutual interference between logistics and commuting flows, and no special intervention is required; the previous adjustment strategy should continue to be implemented.

[0015] Preferably, step four also includes: S43. After the first, second, or third adjustment strategy is issued and executed, collect the ratio of the current green light pedestrian and vehicle flow at the intersection to the previous cycle, and calculate the traffic efficiency improvement value. : ; ; in, Indicates total traffic flow; like This indicates that traffic efficiency has improved, and we proceed to the next step to determine whether the strategy has achieved its optimization goal. like If the current strategy fails to perform satisfactorily, an emergency rollback mechanism will be implemented to restore the previous time slice. The throughput corresponding to −1) The corresponding strategy for -1); S44. When setting the target passage policy improvement value Te for the current policy, calculate the current policy deviation value ‖E‖: ; A preset error tolerance Eth is set. If ||E|| exceeds Eth, it is determined that the deviation has not converged, and the next round of strategy iteration is initiated. If ||E|| does not exceed Eth, the deviation is considered to have converged, thus achieving the optimization objective, and no adjustment is needed.

[0016] A smart traffic control system for cities includes: The multi-source data acquisition module is used to deploy Bluetooth detectors, video cameras, non-motorized vehicle detection coils, and logistics lane flow meters at key intersections in multi-modal traffic areas. It collects real-time data on Bluetooth passenger numbers N(t), single vehicle entry and exit volume S(t), and logistics vehicle flow Q(t) and their geographical location information. Based on the detected network bandwidth and signal stability, it adaptively adjusts the sampling frequency to generate a multimodal dataset. The adaptive sampling processing module periodically detects and records the data distortion status indicators of each channel, including Bluetooth recognition jitter, video image blur, logistics vehicle detection packet loss rate and bandwidth utilization. It constructs and evaluates the multimodal data comprehensive distortion index S, and generates the first throttling sampling strategy or the second sampling strategy. The traffic semantic label modeling module is used to construct a structured set of traffic states L and a corresponding signal timing template library T based on the bus-subway-shared bicycle transfer scenario; and to send L and T to the controllers of each intersection. The overlapping state matching module is used to fuse multimodal datasets obtained from edge nodes or sensing devices, including Bluetooth user numbers N(t), shared bicycle entry and exit volume S(t), video recognition pedestrian density V(t), and logistics vehicle flow Q(t), to generate fused feature vectors. Each fused feature vector is then matched with the standard state features in the traffic state structured set L using cosine similarity to find the matching state label, and the corresponding matching signal timing template T is called. i If the same intersection is in the current time slice When matching multiple state labels, the weight corresponding to the j-th semantic label is used. Calculate the multi-mode overlap index If multiple states overlap, generate the first adjustment strategy; Interference assessment module, if the same intersection is in the current time slice Within the area, there is a simultaneous increase in logistics vehicle traffic and pedestrian / non-motorized vehicle traffic; therefore, the logistics and commuting interference index is calculated. And if there is a risk of interference, a second or third adjustment strategy will be generated. The strategy evaluation and closed-loop optimization module, after the first, second, or third adjustment strategy is issued and executed, collects the ratio of the current green light pedestrian and vehicle flow at the intersection to the previous cycle, and calculates the improvement value of traffic efficiency. The deviation value ‖E‖ from the current strategy is set, and an error tolerance Eth is preset. If the deviation ‖E‖ exceeds Eth, parameter self-learning is triggered, L and T are updated, and iterative optimization is achieved until ‖E‖ converges to the allowable range.

[0017] This invention provides a method and system for intelligent traffic control in cities. It has the following beneficial effects: This invention addresses the problems of slow response and limited control strategies in traditional urban traffic control when dealing with overlapping multi-modal traffic flows (including pedestrian, non-motorized vehicle, and logistics vehicle flows) by introducing multi-source sensing, adaptive sampling, structured semantic modeling, and a composite state recognition mechanism. First, by deploying Bluetooth detectors, video cameras, detection coils, and flow meters at multi-modal traffic intersections, synchronous sensing of pedestrian, bicycle, and logistics vehicle flows is achieved. Combined with adaptive sampling processing under network conditions, data quality and timeliness are ensured. Second, a structured set of traffic states and a corresponding timing template library are constructed, covering various transfer and commuting scenarios, enabling highly compatible and scalable control strategies. By introducing a multi-modal overlap index, when a composite situation such as "peak pedestrian flow + tidal bicycle return + concentrated logistics vehicle crossing" occurs simultaneously at an intersection, the degree of state overlap can be identified in a timely manner, and a collaborative template can be automatically selected to achieve coordinated adjustment of traffic light phases, durations, and priorities, improving control accuracy. Furthermore, by introducing a logistics-commuter interference index model, the intensity of conflict between logistics vehicles and pedestrians and non-motorized vehicles can be assessed in real time. Based on the degree of interference, diverse strategies such as time-limited slowdown, lane reuse, and signal delay are dynamically generated to effectively reduce the risk of logistics traffic disrupting commuting order during peak hours. Finally, by collecting data on changes in traffic efficiency before and after cycles and calculating deviation values, the effectiveness of the strategies is determined in conjunction with a set error tolerance. A closed-loop feedback mechanism is constructed to trigger a parameter self-learning process, dynamically updating semantic labels and timing templates, enabling the control system to continuously optimize and quickly adapt to changes in traffic conditions. In summary, this invention has significant effects on improving traffic efficiency, alleviating concurrent congestion of multiple flows, and reducing the risk of pedestrian-vehicle conflicts. It overcomes the shortcomings of traditional solutions, such as low recognition accuracy, large response delay, and inability to handle complex situations, and has good practical value and promising prospects for promotion. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating the steps of a smart traffic control method for cities according to the present invention; Figure 2 This is a schematic diagram of a smart traffic control system for cities according to the present invention. Detailed Implementation

[0019] 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.

[0020] Example 1 Please see Figure 1This invention provides a smart traffic control method for cities, comprising the following steps: Step 1: Multi-source traffic data acquisition and adaptive preprocessing: Deploy Bluetooth detectors, video cameras, non-motorized vehicle detection loops, and logistics lane flow meters at key intersections in multi-modal traffic convergence areas to collect real-time data on Bluetooth passenger numbers N(t), single vehicle entry and exit volume S(t), and logistics vehicle flow Q(t) and their geographical location information; adaptively adjust the sampling frequency based on the detected network bandwidth and signal stability to generate a multi-modal dataset; Step 2: Establish and distribute standard traffic state semantics and template library: Based on the bus-subway-shared bicycle transfer scenario, construct a structured set of traffic states L and a corresponding signal timing template library T; distribute L and T to the controllers of each intersection. Step 3: Real-time Multimodal Fusion Recognition and Traffic State Detection: Normalize and time-align the multimodal dataset, project it onto a unified feature dimension to obtain the fused feature vector F(τ); perform similarity matching between F(τ) and the standard semantic label L and template library T to construct a multimodal overlap index. And assess, if >0.6, generate the first adjustment strategy; Step 4: Generate functional traffic control instructions and perform closed-loop iterative optimization: Construct a logistics and commuting interference index. The system then evaluates and generates second and third adjustment strategies, while simultaneously collecting real-time data on the improvement in intersection traffic efficiency. The deviation value ‖E‖ from the current strategy is set, and an error tolerance Eth is preset. If the deviation ‖E‖ exceeds Eth, parameter self-learning is triggered, L and T are updated, and iterative optimization is achieved until ‖E‖ converges to the allowable range.

[0021] This invention addresses the problems of slow response and limited control strategies in traditional urban traffic control when dealing with overlapping multi-modal traffic flows (including pedestrian, non-motorized vehicle, and logistics vehicle flows) by introducing multi-source sensing, adaptive sampling, structured semantic modeling, and a composite state recognition mechanism. First, by deploying Bluetooth detectors, video cameras, detection coils, and flow meters at multi-modal traffic intersections, synchronous sensing of pedestrian, bicycle, and logistics vehicle flows is achieved. Combined with adaptive sampling processing under network conditions, data quality and timeliness are ensured. Second, a structured set of traffic states and a corresponding timing template library are constructed, covering various transfer and commuting scenarios, enabling highly compatible and scalable control strategies. By introducing a multi-modal overlap index, when a composite situation such as "peak pedestrian flow + tidal bicycle return + concentrated logistics vehicle crossing" occurs simultaneously at an intersection, the degree of state overlap can be identified in a timely manner, and a collaborative template can be automatically selected to achieve coordinated adjustment of traffic light phases, durations, and priorities, improving control accuracy. Furthermore, by introducing a logistics-commuter interference index model, the intensity of conflict between logistics vehicles and pedestrians and non-motorized vehicles can be assessed in real time. Based on the degree of interference, diverse strategies such as time-limited slowdown, lane reuse, and signal delay are dynamically generated to effectively reduce the risk of logistics traffic disrupting commuting order during peak hours. Finally, by collecting data on changes in traffic efficiency before and after cycles and calculating deviation values, the effectiveness of the strategies is determined in conjunction with a set error tolerance. A closed-loop feedback mechanism is constructed to trigger a parameter self-learning process, dynamically updating semantic labels and timing templates, enabling the control system to continuously optimize and quickly adapt to changes in traffic conditions. In summary, this invention has significant effects on improving traffic efficiency, alleviating concurrent congestion of multiple flows, and reducing the risk of pedestrian-vehicle conflicts. It overcomes the shortcomings of traditional solutions, such as low recognition accuracy, large response delay, and inability to handle complex situations, and has good practical value and promising prospects for promotion.

[0022] Example 2 This embodiment is an explanation of Embodiment 1. Specifically, the multi-source traffic data collection in step one includes the following sub-steps: S11. Install the following sensing devices at key intersections in multimodal traffic convergence areas: Bluetooth detectors are used to sense the density of people entering and exiting subway stations, bus stations, and intersection transfer nodes in real time, and to identify the number of people N(t) via Bluetooth devices. Video cameras are used to be installed on high-level road poles or platform areas to collect video streams for identifying situations where pedestrians and vehicles are mixed and for recognizing traffic behavior characteristics. Non-motorized vehicle detection loops are laid near the entrance of the non-motorized vehicle lane and the zebra crossing to count the number of shared bicycles and electric vehicles passing through, S(t). Logistics vehicle flow monitors, including radar, infrared, or geomagnetic flow meters, are used to be installed at intersections and logistics channel entrances to collect logistics vehicle flow data Q(t). Logistics lane flow meters are used to obtain freight vehicle flow. S12, and configure network time synchronization modules in edge nodes, including roadside computing devices, to align the timestamps of data from various sensing devices through the master-slave clock mechanism PTP protocol.

[0023] The adaptive preprocessing in step one includes the following sub-steps: S13. Periodically detect and record the data distortion indicators of each channel, including Bluetooth recognition jitter, video image blur, and packet loss rate and bandwidth utilization of logistics vehicle detection; fuse the quality indicators of each modal sensing signal in a weighted manner to construct a multimodal data comprehensive distortion index S, the calculation formula of which is: ; In the formula, Indicates Bluetooth density volatility. Indicates the frame rate fluctuation of video images. For network data packet loss rate, Indicates the jitter of the network communication link. , , and It is a constant, and , Indicates according to , , and The weights are used to calculate the overall distortion index S of the multimodal data; The above weights , , and The weighting is primarily based on the influence and criticality of each channel in the overall data quality. For example, Bluetooth density fluctuation reflects the stability of pedestrian recognition, video image frame rate fluctuation affects the accuracy of human and vehicle behavior recognition, and network packet loss rate and communication link jitter determine data integrity and real-time performance. Weight values ​​can be set based on historical sampling performance, key application requirements, and sensitivity to typical faults. An example setting is: , , and These correspond to 0.3, 0.3, 0.2, and 0.2 respectively, indicating that Bluetooth and video signals have a greater impact on the overall distortion, and therefore are given relatively higher weights. The system can also dynamically adjust these weights through continuous learning to adapt to different scenarios and time periods.

[0024] S14. A first threshold Sth1 and a second lower threshold Sth2 are preset. If the multimodal data comprehensive distortion index S exceeds the first threshold Sth1, it indicates that the current data transmission command is unqualified, and there is a risk of network congestion or sensor data redundancy. The first throttling sampling strategy is then executed, including: The frame rate of the video image channel is reduced by 25% to 50%, from 30fps to 20fps or 15fps; the image resolution is reduced by one level (e.g., from 1080p to 720p); at the same time, video clips of target areas (such as zebra crossings and lanes) are prioritized; the polling cycle of the Bluetooth identification channel is extended by 30% to 60%; for example, from scanning once every 5 seconds to once every 8 seconds or every 10 seconds, to reduce signal collisions and communication pressure; the data upload interval of non-motorized vehicle detectors / logistics detectors is extended by 20% to 40%, and an edge caching mechanism is enabled, uploading only when the detection result changes exceed a threshold, in order to compress the total data traffic; when S exceeds Sth1, the following transmission order is followed, including: video channel compression priority > Bluetooth scanning priority > detector reporting priority; S15. If the multimodal data comprehensive distortion index S is lower than the second lower limit threshold Sth2, it indicates that the current data transmission command is qualified and resources are abundant, meeting the conditions for executing high-frequency, high-fidelity sampling strategies. Therefore, the second sampling strategy is triggered, including: the system will use this "marginal time" to restore or optimize the data quality of various sensors to obtain more refined traffic dynamic features; the image frame rate of the video image channel will be restored to a proportional 30fps, and the image resolution will be increased from 720p to 1080p; the video sampling rate in areas with mixed pedestrian and vehicle traffic will be increased by 10%; the polling cycle of the Bluetooth identification channel will be shortened by 15% to 30%, for example, from 10 seconds to 7 seconds or 5 seconds, to enhance the real-time perception of pedestrian flow fluctuations; the data of non-motorized vehicle detectors / logistics detectors will be restored to the normal upload frequency, and a short-cycle fluctuation value early warning mechanism will be enabled, especially when the morning and evening peak hours are approaching, the reporting frequency will be increased by about 10% to 25%; if the shared bicycle entry and exit volume S(t) changes drastically by ±40%, the sampling frequency priority of the non-motorized vehicle lane will be temporarily increased, while other channels will remain unchanged. If the multimodal data overall distortion index S falls within the range of the first threshold Sth1 and the second lower threshold Sth2, it indicates that the current data transmission instruction is qualified and no adjustment is required. The source of the first threshold Sth1 (upper limit): This value is the critical tolerance upper limit value calculated by the system through sampling distortion indicators (such as Bluetooth fluctuations, image blurring, packet loss rate, etc.) in historical operating data. It indicates that once the degree of data distortion exceeds this value, there is a significant risk of abnormality in the system's data acquisition link. It is necessary to compress the sampling and reduce the load in time to prevent system crash or loss of critical data.

[0025] The source of the second lower limit threshold Sth2 (lower limit): This value represents the lowest level of distortion that the system can accept during normal operation. That is, if the distortion level is lower than this value, it means that the current network and equipment are in good condition and have the conditions to perform high-frequency, high-fidelity data sampling. The sampling quality can be actively improved to achieve refined traffic perception.

[0026] S16. After the first throttling sampling strategy or the second sampling strategy is executed, a multimodal dataset is generated.

[0027] In this embodiment, an adaptive preprocessing mechanism is introduced to address issues such as data quality fluctuations, network congestion, and waste of sensing resources during multi-source sensing data transmission in urban multi-modal traffic environments. On one hand, various sensing devices, including Bluetooth detectors, video cameras, non-motorized vehicle detection coils, and logistics vehicle flow meters, are employed to achieve full-element sensing coverage of pedestrian, non-motorized, and logistics vehicle flows, ensuring the breadth and granularity of multimodal traffic data collection. On the other hand, by constructing multiple quality indicators such as Bluetooth recognition jitter, video image blur, logistics detection packet loss rate, and link jitter, a multimodal data comprehensive distortion index S is generated, which can dynamically evaluate the data channel's operating status. Based on this evaluation result, the system can execute a first throttling sampling strategy when there is network congestion or high sensor load, effectively compressing redundant traffic, reducing communication pressure, and ensuring the priority of core data transmission. During periods of ample network coverage and stable signals, the system can proactively trigger a second sampling strategy to increase the image frame rate and Bluetooth scanning frequency, enhancing the response sensitivity to sudden changes in pedestrian and vehicle flow. By finely controlling the frame rate, resolution, polling cycle, and upload frequency of different sensing channels, the system has the ability to adjust the data acquisition accuracy according to time and location, thereby significantly enhancing the transmission stability, sensing accuracy, and system robustness of multi-source traffic data, and providing high-quality data support for subsequent overlapping state detection and adjustment strategy generation.

[0028] Example 3 This embodiment is an explanation based on embodiment 1. Specifically, the steps for constructing the structured set L of traffic states in step two are as follows: S21. Decompose the typical traffic scenarios in the bus-subway-shared bicycle transfer process, divide the flow of people, vehicles, signal control, and logistics lane occupation into a "state-trigger-response" triplet, and collect at least 30 days of historical multimodal traffic data, including Bluetooth number of people N(t), bicycle entry and exit volume S(t), and logistics vehicle flow Q(t). S22. Perform sliding window processing on each type of traffic data, specifically including Bluetooth passenger count N(t), single vehicle entry / exit count S(t), and logistics vehicle flow Q(t). Set the time window to 5 minutes and the sliding step size to 1 minute. Extract each time segment as a state unit, with the expression as follows: ; This represents the data set of the k-th state segment, also known as the k-th time window; it is a set of data segments used to calculate growth rate, trend, or volatility indicators. This represents the traffic data collected at time t, including any one or more vectors among Bluetooth users N(t), single vehicle entry and exit S(t), and logistics vehicle flow Q(t); Indicates the current window The time range starts from the beginning time. Start, duration is All data falling within this time frame belongs to this state segment; S23. For each intersection in each time segment, calculate the first derivative of the number of pedestrians via Bluetooth at time t: ; express The absolute value of the change in the number of people over a period of time; divided by This is to calculate the average rate of change, in units of people per minute; S24. For each intersection in each time segment, calculate the second derivative of the number of pedestrians via Bluetooth at time t: ; First, calculate the difference of the first derivatives of the two consecutive time windows, i.e., the change in the growth rate, and then divide by... Acceleration is obtained; This represents finding the first derivative of the number of people with respect to time. It represents the second derivative, which measures the "change" in the rate of change—that is, acceleration; S25. Calculate the total flow of shared bicycles S(t) at time t as: S(t) = Sin(t) + Sout(t); Sin(t) represents the number of shared bicycles entering at time t, and Sout(t) represents the number of shared bicycles leaving at time t. S26. Calculate the rate of change of logistics vehicle traffic within the time window using the first derivative of logistics vehicle traffic: ; This represents the flow of logistics vehicles collected at time t.

[0029] S27, Preset first threshold Second threshold and the third threshold If, within three consecutive time windows, i.e., the current time slice τ∈{t,t−Δt,t−2Δt}, the following conditions are met: The first derivative of the number of pedestrians via Bluetooth satisfies

[0030] The second derivative of the number of pedestrians via Bluetooth satisfies Generate a "high-density transfer" status label for this time period [t−2Δt, t]. If the total flow of shared bicycles S(t) is relative to the maximum capacity Cbike of the road segment, then: ; When the duration of this condition exceeds the minimum duration threshold of peak single-vehicle conditions, a "single-vehicle tidal return" status label is generated. If the first derivative of the logistics vehicle flow rate satisfies: ; If the corresponding vehicles are mainly concentrated at adjacent intersections less than 2km apart, a "static intersection congestion" status label will be generated. First threshold Second threshold and the third threshold The specific values ​​are set based on factors such as historical urban traffic data, road capacity, and sensor accuracy. Generally, the pedestrian flow rate threshold can be set to no less than 10 people per minute. When the number of pedestrians at a certain intersection changes by more than this value per unit time, it indicates a significant change in pedestrian flow. The pedestrian flow acceleration threshold is usually set at more than 5 people per minute squared to determine whether the pedestrian flow growth trend is accelerating, reflecting obvious aggregation or tidal phenomena. The shared bicycle flow ratio threshold is generally set at 85% of the maximum capacity. When the bicycle flow reaches this ratio, it indicates that the road is in a high-density state, with a risk of backflow or congestion. These thresholds can be dynamically adjusted and optimized based on the specific traffic characteristics and collected data of a city.

[0031] S29. Organize the semantic labels of the status labels within each time interval that meets the conditions into a structured set L of traffic status. , to The semantic label represents the first to the nth semantic label that meets the conditions; the structured semantic label set L contains multiple semantic labels, each label is accompanied by a set of standard state feature vectors generated by summarizing historical multimodal traffic data, which are used to characterize the standard state features of the semantic label in multiple dimensions such as Bluetooth user changes, single vehicle tidal degree, logistics vehicle fluctuation and mixed traffic degree. Each semantic tag includes: tag type, time interval, corresponding intersection code or geographic location information, and feature variable values ​​supporting tag judgment; each semantic tag in the traffic state structured set L matches one or more signal timing templates. This establishes a mapping relationship and constructs a corresponding signal timing template library T. , to This represents the signal timing template from the first signal timing template to the nth signal timing template that meets the conditions for each semantic tag; Signal timing template This includes parameters such as green light duration, red light delay, phase switching sequence, and priority parameters for specific lanes.

[0032] In this embodiment, a structured set of traffic states, L, and its mapped signal timing template library, T, are constructed, significantly improving the ability to finely characterize and respond to complex dynamic states in urban multimodal traffic scenarios. On the one hand, by adopting a "state-trigger-response" triplet modeling approach, combined with at least 30 days of historical Bluetooth ridership, vehicle entry and exit volume, and logistics vehicle flow data, the system can accurately identify typical complex traffic states such as "high-density transfers," "tidal return of single vehicles," and "static congested intersections." Through sliding time window processing, multidimensional derivative calculations, and threshold logic judgment, the evolution trend, fluctuation amplitude, and spatial coupling relationship of different traffic factors are extracted, giving the state labels stable structured semantics. On the other hand, by mapping each semantic label to a signal timing template, the system enables the signal control strategy to have "label-based invocation" and "scenario-based matching" capabilities, significantly improving the adaptability and intelligence level of signal timing, especially suitable for highly dynamic and complex intersection scenarios such as dense pedestrian flow, concentrated logistics, and non-motorized vehicle tidal mixing. Through this structured modeling method and template mapping mechanism, the system can not only respond quickly to sudden or superimposed traffic situations.

[0033] Example 4 This embodiment is an explanation based on embodiment 3. Specifically, the steps for deploying and calling the template library in step two are as follows: The constructed traffic state structure set L and the corresponding signal timing template library T are encapsulated into a unified data structure to generate the first control index table, specifically in SON or XML format, which facilitates network transmission and embedded decoding; the data is pushed to the central control platform via an edge router using serial bus RS485 or CAN, network protocols IP / TCP, MQTT, or WebSocket. The centralized control platform reads the first control index table and extracts the k-th state segment of each sampling window. If the tag condition is met, and tag L is matched, immediately invoke the bound signal timing template. ,include: Change the green light duration, start the left turn phase in advance, give priority to non-motorized vehicles, and coordinate the start of traffic lights.

[0034] In this embodiment, by encapsulating the structured traffic state set L and the signal timing template library T into a unified data structure (such as JSON or XML format), data standardization and structuring are achieved, facilitating efficient network transmission and rapid parsing by embedded devices. Multiple communication protocols (such as RS485, CAN, IP / TCP, MQTT, and WebSocket) are used to push the control index table to the centralized control platform via an edge router, ensuring the stability and real-time performance of data transmission. The centralized control platform can parse the index table in real time, quickly identify whether the current traffic state segment meets the preset label conditions, and accurately call the corresponding signal timing template to dynamically adjust the green light duration, advance the left turn phase activation, prioritize the passage of non-motorized vehicles, and activate the intersection traffic light coordination strategy. This mechanism effectively improves the response speed and control accuracy of traffic signal control, enhances the system's adaptability to complex, multi-state superimposed traffic flows, thereby alleviating intersection congestion and improving traffic efficiency and safety.

[0035] Example 5 This embodiment is an explanation based on Embodiment 1. Specifically, the steps for real-time multimodal overlap fusion recognition in step three are as follows: S31. A multimodal dataset obtained from edge nodes or sensing devices, including Bluetooth user count N(t), shared bicycle entry and exit volume S(t), video recognition pedestrian density V(t), and logistics vehicle flow Q(t), is executed as follows: Time alignment processing ensures that data from each channel corresponds to the same state segment at the same time t. Feature normalization processing normalizes various types of data to a unified interval [0,1]. Projecting onto a unified feature space generates a fused feature vector: S32. Perform cosine similarity matching between each fused feature vector and the standard state features in the traffic state structure set L to find the matching state label, and call the corresponding matching signal timing template T. i ; If multiple tags simultaneously meet the matching conditions, they are recorded as a multi-tag state set. Real-time collection and fusion of multi-source traffic data to form the current time slice Feature vector Calculate cosine similarity : ; If there exists an 's'-th element that satisfies the condition, then label it. satisfy If the value is ≥0.85, it is considered a "successful match", and then the matching signal timing template T is called accordingly. i .

[0036] S33. If the same intersection is in the current time slice When matching multiple state labels, the weight corresponding to the j-th semantic label is used. ∈[0,1], calculate the multi-mode overlap index : ; in, This represents the total number of semantic tags that were successfully matched. Indicates the first The weight of each semantic label; No. Weight of each semantic label The source of the label is a weighted result of the identification confidence and regulatory impact of the label in similar spatiotemporal contexts within historical traffic data. Specifically, the higher the frequency with which a label is identified by the system in similar scenarios, the more common it is, and the greater its weight. Historically, the more significant the improvement effect of the adjustment strategy corresponding to this label on traffic, the greater its weight. For example, when "pedestrian-vehicle conflict" affects safety or traffic flow more than "bicycle return," it will be given a higher weight. Labels near morning and evening rush hours, commercial areas, and subway entrances are more likely to be effective and will also be given higher weight.

[0037] The weight of high-density transfer status (such as the morning rush hour crowds at subway entrances) is set to 0.42; the weight of bicycle tidal return status (such as the concentrated return of shared bicycles) is set to 0.28; the weight of logistics vehicle flow surge status (such as the concentrated entry of freight trucks into business districts in the morning) is set to 0.30; and the weight of semantic tags is 1.

[0038] If the multi-mode overlap index A value greater than 0.6 indicates multiple overlapping states, triggering the first adjustment strategy. This strategy prioritizes the use of a "multi-mode collaborative template," extending the green light duration for pedestrian lights, non-motorized vehicle lights, and freight lights by 10%–15%, 8%–12%, and 5%, respectively. It also advances the left-turn phase by 5–7 seconds to reduce intersection waiting conflicts. Non-motorized vehicles are given priority as first-level right-of-way, ensuring their passage during peak pedestrian and bicycle traffic periods. Furthermore, a coordination strategy with adjacent intersection traffic lights is implemented to improve tidal flow traffic by at least 18%. If the multi-mode overlap index A value <0.3 indicates that the current traffic condition is clearly defined and has a single label, requiring maintenance of the previous control cycle without the need for advance control. 0.7 and 0.3 are reasonable threshold ranges determined through repeated data-driven debugging and optimization, based on actual traffic flow patterns and system performance indicators, aiming to achieve balanced response and stable operation of the intelligent transportation system.

[0039] In this embodiment, by performing time alignment and normalization on the multimodal dataset, effective fusion of traffic data from different sources is achieved, ensuring the timeliness and consistency of the fused feature vector. Based on cosine similarity, the fused feature vector is accurately matched with standard traffic states, enabling sensitive identification of the current traffic state. When multiple state labels match simultaneously, a weighted multimodal overlap index is calculated to scientifically quantify the degree of state overlap. For complex traffic situations with high overlap indices, a multimodal collaborative template is prioritized, reasonably extending the green light duration for pedestrian, non-motorized vehicle, and freight lights, initiating left-turn phases in advance, and increasing the priority of non-motorized vehicles. Coordination with traffic lights at adjacent intersections significantly reduces waiting times and conflicts, improving traffic flow during peak hours by at least 18%. This method not only enhances the system's ability to identify complex multimodal traffic states and its response flexibility but also effectively alleviates congestion caused by the interweaving of pedestrians, vehicles, and goods, improving traffic efficiency and safety, and ensuring intelligent and precise traffic control.

[0040] Example 6 This embodiment is an explanation of Embodiment 1. Specifically, step four includes: S41. If the same intersection is in the current time slice Within the area, there is a simultaneous increase in logistics vehicle traffic and pedestrian / non-motorized vehicle traffic; therefore, the logistics and commuting interference index is calculated. : ; in, This indicates the capacity of the mixed flow channel, which is preset based on the intersection's geographical structure, number of lanes, and width of pedestrian and non-motorized vehicle lanes. , and Each is a current time slice Bluetooth user count, bicycle entry and exit volume, and logistics vehicle traffic; S42, if A value of ≥1.2 indicates that the flow of logistics vehicles has posed a first risk of interference to commuter / non-motorized vehicle traffic, generating a second adjustment strategy, including: delaying the release phase of logistics vehicles, reducing the release time per cycle by 20% to 30%; designating 30% of the original freight lanes as "time-limited slow-moving lanes" or "temporary U-turn lanes" for a duration of no less than 15 minutes; increasing the green light duration by 12% to 20%; and responding to pedestrian crossing signals 3 seconds earlier. If the interference continues for more than 3 cycles, with each cycle lasting 5 minutes, then a structural traffic restriction mode will be implemented, requiring logistics vehicles at the intersection to detour via the auxiliary road. If 0.8≤ ≤1.2 indicates a second risk of interference between pedestrian and vehicle traffic, generating a third adjustment strategy, including: adjusting the release delay of logistics vehicles by 10% to 15%; extending the green light switching buffer by 3 to 5 seconds to prevent overlapping conflicts between pedestrians and vehicles; releasing shared bicycles 5 seconds earlier with the green light; and triggering pop-up guidance to avoid logistics channels in densely populated shared bicycle sections. like A value of <0.8 indicates that there is no risk of mutual interference between logistics and commuting flows, and no special intervention is required; the previous adjustment strategy should continue to be implemented.

[0041] In this embodiment, step four scientifically identifies the risk level of concurrent enhanced interference between logistics vehicle flow, pedestrian flow, and non-motorized vehicle flow by calculating the logistics and commuting interference index in real time, thus achieving precise risk-level response. For high-risk (≥1.2) states, the system effectively delays the release of logistics vehicles, reasonably reduces the release time, and flexibly divides some freight lanes into time-limited slow-moving or U-turn lanes to reduce the obstruction of commuting flow by logistics, significantly improves the green light response speed of pedestrian crossings, and activates structural traffic restrictions under continuous interference conditions to ensure smooth and safe commuting. For medium-risk (0.8–1.2) states, by appropriately delaying the release of logistics and extending the green light switching buffer interval, conflicts between pedestrians and vehicles are effectively avoided. Simultaneously, the passage of shared bicycles is coordinated, and pop-up guidance is activated to achieve dynamic diversion and avoidance. For low-risk (<0.8) states, the current strategy is maintained to avoid unnecessary intervention and ensure stable system operation. This tiered response mechanism enhances the flexibility and intelligence of traffic control, reduces congestion and safety hazards caused by the intermingling of logistics and commuter flows, effectively ensures smooth and safe urban traffic, and promotes the coordinated development of multimodal transportation.

[0042] Example 7 This embodiment is based on the explanation in Embodiment 1. Specifically, step four also includes: S43. After the first, second, or third adjustment strategy is issued and executed, collect the ratio of the current green light pedestrian and vehicle flow at the intersection to the previous cycle, and calculate the traffic efficiency improvement value. : ; ; in, Indicates total traffic flow; like This indicates that traffic efficiency has improved, and we proceed to the next step to determine whether the strategy has achieved its optimization goal. like If the current strategy fails to perform satisfactorily, an emergency rollback mechanism will be implemented to restore the previous time slice. The throughput corresponding to −1) The corresponding strategy for -1); S44. When setting the target passage policy improvement value Te for the current policy, calculate the current policy deviation value ‖E‖: ; A preset error tolerance Eth is set. If ||E|| exceeds Eth, it is determined that the deviation has not converged, and the next round of strategy iteration is initiated. If ||E|| does not exceed Eth, the deviation is considered to have converged, thus achieving the optimization objective, and no adjustment is needed.

[0043] In this embodiment, step four involves comparing real-time pedestrian and vehicle traffic flow under green light conditions with data from the previous period to scientifically assess the efficiency improvement of the current traffic adjustment strategy. When the traffic efficiency fails to meet expectations, the system activates an emergency rollback mechanism to restore the effective strategy from the previous time slice, ensuring smooth and stable traffic flow. Combining the preset target traffic improvement value and error tolerance, the strategy deviation is dynamically calculated. If the deviation exceeds a threshold, parameter self-learning and iterative strategy adjustment are automatically triggered, achieving continuous optimization of the signal timing scheme. This closed-loop feedback mechanism enhances the system's adaptability and response speed to complex traffic conditions, effectively preventing a disconnect between the strategy and actual operation, ensuring the efficiency and stability of traffic control at intersections, and improving overall traffic efficiency and safety.

[0044] Example 8 A smart traffic control system for cities, please refer to Figure 2 ,include: The multi-source data acquisition module is used to deploy Bluetooth detectors, video cameras, non-motorized vehicle detection coils, and logistics lane flow meters at key intersections in multi-modal traffic areas. It collects real-time data on Bluetooth passenger numbers N(t), single vehicle entry and exit volume S(t), and logistics vehicle flow Q(t) and their geographical location information. Based on the detected network bandwidth and signal stability, it adaptively adjusts the sampling frequency to generate a multimodal dataset. The adaptive sampling processing module periodically detects and records the data distortion status indicators of each channel, including Bluetooth recognition jitter, video image blur, logistics vehicle detection packet loss rate and bandwidth utilization. It constructs and evaluates the multimodal data comprehensive distortion index S, and generates the first throttling sampling strategy or the second sampling strategy. The traffic semantic label modeling module is used to construct a structured set of traffic states L and a corresponding signal timing template library T based on the bus-subway-shared bicycle transfer scenario; and to send L and T to the controllers of each intersection. The overlapping state matching module is used to fuse multimodal datasets obtained from edge nodes or sensing devices, including Bluetooth user numbers N(t), shared bicycle entry and exit volume S(t), video recognition pedestrian density V(t), and logistics vehicle flow Q(t), to generate fused feature vectors. Each fused feature vector is then matched with the standard state features in the traffic state structured set L using cosine similarity to find the matching state label, and the corresponding matching signal timing template T is called. i If the same intersection is in the current time slice When matching multiple state labels, the weight corresponding to the j-th semantic label is used. Calculate the multi-mode overlap index If multiple states overlap, generate the first adjustment strategy; Interference assessment module, if the same intersection is in the current time slice Within the area, there is a simultaneous increase in logistics vehicle traffic and pedestrian / non-motorized vehicle traffic; therefore, the logistics and commuting interference index is calculated. And if there is a risk of interference, a second or third adjustment strategy will be generated. The strategy evaluation and closed-loop optimization module, after the first, second, or third adjustment strategy is issued and executed, collects the ratio of the current green light pedestrian and vehicle flow at the intersection to the previous cycle, and calculates the improvement value of traffic efficiency. The deviation value ‖E‖ from the current strategy is set, and an error tolerance Eth is preset. If the deviation ‖E‖ exceeds Eth, parameter self-learning is triggered, L and T are updated, and iterative optimization is achieved until ‖E‖ converges to the allowable range.

[0045] This system achieves comprehensive perception of pedestrian, non-motorized vehicle, and logistics traffic at key intersections through a multi-source data acquisition module, ensuring data timeliness and accuracy. An adaptive sampling processing module automatically adjusts sampling frequency and precision based on real-time data quality, improving system robustness and resource utilization efficiency under different network environments. A traffic semantic labeling and template library module, combined with multi-modal traffic transfer scenarios, constructs structured traffic state and timing templates, enabling accurate state recognition and signal control scheme matching. An overlapping state matching module, by fusing multi-modal data and calculating multi-modal overlap indices, effectively identifies complex overlapping traffic states, ensuring the scientific rationality of adjustment strategies. An interference assessment module dynamically adjusts strategies for concurrent logistics and commuter flows, alleviating conflicts and congestion between logistics vehicles and pedestrian / vehicle traffic. A strategy evaluation and closed-loop optimization module enables real-time monitoring and deviation feedback of control effects, automatically triggering parameter self-learning and strategy iterative optimization, ensuring the system adapts to complex and changing traffic environments and improving intersection efficiency and traffic safety. The overall system possesses high intelligence, real-time response, and adaptive optimization capabilities, significantly improving urban traffic flow and travel experience.

[0046] The threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by those skilled in the art for each set of sample data; as long as it does not affect the ratio between the parameter and the quantized value, it is acceptable.

[0047] The above formulas are all derived from software simulation using a large amount of data and are selected to be close to the actual values. The coefficients in the formulas are set by those skilled in the art according to the actual situation. The above description is only a preferred embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any equivalent substitutions or changes made by those skilled in the art within the technical scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the protection scope of the present invention.

Claims

1. A smart traffic control method for cities, characterized in that, Includes the following steps: Step 1: Multi-source traffic data acquisition and adaptive preprocessing: Deploy Bluetooth detectors, video cameras, non-motorized vehicle detection loops, and logistics lane flow meters at key intersections in multi-modal traffic convergence areas to collect real-time data on Bluetooth passenger numbers N(t), single vehicle entry and exit volume S(t), and logistics vehicle flow Q(t) and their geographical location information; adaptively adjust the sampling frequency based on the detected network bandwidth and signal stability to generate a multi-modal dataset; Step 2: Establish and distribute standard traffic state semantics and template library: Based on the bus-subway-shared bicycle transfer scenario, construct a structured set of traffic states L and a corresponding signal timing template library T; distribute L and T to the controllers of each intersection. Step 3: Real-time Multimodal Fusion Recognition and Traffic State Detection: Normalize and time-align the multimodal dataset, project it onto a unified feature dimension to obtain the fused feature vector F(τ); perform similarity matching between F(τ) and the standard semantic label L and template library T to construct a multimodal overlap index. And assess, if >0.6, generate the first adjustment strategy; Step 4: Generate functional traffic control instructions and perform closed-loop iterative optimization: Construct a logistics and commuting interference index. The system then evaluates and generates second and third adjustment strategies, while simultaneously collecting real-time data on the improvement in intersection traffic efficiency. The deviation value ‖E‖ from the current strategy is set, and an error tolerance Eth is preset. If the deviation ‖E‖ exceeds Eth, parameter self-learning is triggered, L and T are updated, and iterative optimization is achieved until ‖E‖ converges to the allowable range.

2. The intelligent traffic control method for cities according to claim 1, characterized in that, The multi-source traffic data collection in step one includes the following sub-steps: S11. Install Bluetooth detectors, video cameras, non-motorized vehicle detection coils, and logistics vehicle flow monitors at key intersections in multi-modal traffic areas to collect data on Bluetooth users N(t), single vehicle entry and exit S(t), and logistics vehicle flow Q(t). S12, and configure network time synchronization modules in edge nodes, including roadside computing devices, to align the timestamps of data from various sensing devices through the master-slave clock mechanism PTP protocol.

3. The intelligent traffic control method for cities according to claim 2, characterized in that, The adaptive preprocessing in step one includes the following sub-steps: S13. Periodically detect and record the data distortion status indicators of each channel, including Bluetooth recognition jitter, video image blur, logistics vehicle detection packet loss rate and bandwidth utilization; integrate the quality indicators of each modal sensing signal in a weighted manner to construct a multimodal data comprehensive distortion index S; S14. A first threshold Sth1 and a second lower threshold Sth2 are preset. If the multimodal data comprehensive distortion index S exceeds the first threshold Sth1, it indicates that the current data transmission command is unqualified, and there is a risk of network congestion or sensor data redundancy. The first throttling sampling strategy is then executed, including: The image frame rate of the video image channel is reduced by 25% to 50%, from 30fps to 20fps or 15fps; the polling cycle of the Bluetooth identification channel is extended by 30% to 60%; the data upload interval of the non-motorized vehicle detector / logistics detector is extended by 20% to 40%, and an edge caching mechanism is enabled, uploading only when the detection result changes exceed the threshold, in order to compress the total data traffic; when S exceeds Sth1, the following transmission order is followed, including: video channel compression priority > Bluetooth scanning priority > detector reporting priority; S15. If the multimodal data comprehensive distortion index S is lower than the second lower limit threshold Sth2, it indicates that the current data transmission command is qualified and resources are abundant, and the conditions for executing high-frequency, high-fidelity sampling strategies are met. Therefore, the second sampling strategy is triggered, including: restoring the image frame rate of the video image channel to a proportional 30fps, and increasing the image resolution from 720p to 1080p; prioritizing an increase of 10% in the video sampling rate of mixed pedestrian and vehicle areas; shortening the polling cycle of the Bluetooth identification channel by 15% to 30%; restoring the data upload frequency of non-motorized vehicle detectors / logistics detectors to the normal upload frequency, and enabling a short-cycle fluctuation value early warning mechanism, especially when the morning and evening peak hours are approaching, increasing the reporting frequency by about 10% to 25%; if the shared bicycle entry and exit volume S(t) changes drastically by ±40%, the sampling frequency priority of the non-motorized vehicle lane channel will be temporarily increased, while other channels will remain unchanged. If the multimodal data overall distortion index S falls within the range of the first threshold Sth1 and the second lower threshold Sth2, it indicates that the current data transmission instruction is qualified and no adjustment is required. S16. After the first throttling sampling strategy or the second sampling strategy is executed, a multimodal dataset is generated.

4. The intelligent traffic control method for cities according to claim 1, characterized in that, The specific steps for constructing the structured set L of traffic states in step two are as follows: S21. Decompose the typical traffic scenarios in the bus-subway-shared bicycle transfer process, divide the pedestrian flow, vehicle flow, signal control, and logistics lane occupation operation into "state-trigger-response" triplet, and collect at least 30 days of historical multimodal traffic data, including Bluetooth number of people N(t), bicycle entry and exit volume S(t), and logistics vehicle flow Q(t). S22. Perform sliding window processing on each type of traffic data, specifically including Bluetooth passenger count N(t), single vehicle entry / exit count S(t), and logistics vehicle flow Q(t). Set the time window to 5 minutes and the sliding step size to 1 minute. Extract each time segment as a state unit, with the expression as follows: ; This represents the k-th state segment; This represents the traffic data collected at time t, including any one or more vectors among Bluetooth users N(t), single vehicle entry and exit S(t), and logistics vehicle flow Q(t); Indicates the current window The time range starts from the beginning time. Start, duration is All data falling within this time frame belongs to this state segment; S23. For each intersection in each time segment, calculate the first derivative of the number of pedestrians via Bluetooth at time t: ； express The absolute value of the change in the number of people over a period of time; divided by This is to calculate the average rate of change, in units of people per minute; S24. For each intersection in each time segment, calculate the second derivative of the number of pedestrians via Bluetooth at time t: ； First, calculate the difference of the first derivatives of the two consecutive time windows, i.e., the change in the growth rate, and then divide by... Acceleration is obtained; S25. Calculate the total flow of shared bicycles S(t) at time t as: S(t) = Sin(t) + Sout(t); Sin(t) represents the number of shared bicycles entering at time t, and Sout(t) represents the number of shared bicycles leaving at time t. S26. Calculate the rate of change of logistics vehicle traffic within the time window using the first derivative of logistics vehicle traffic: ； This represents the flow of logistics vehicles collected at time t; S27, Preset first threshold Second threshold and the third threshold ; If, within three consecutive time windows, i.e., the current time slice τ∈{t,t−Δt,t−2Δt}, the following condition is met: The first derivative of the number of pedestrians via Bluetooth satisfies ; The second derivative of the number of pedestrians via Bluetooth satisfies Generate a "high-density transfer" status label for this time period [t−2Δt, t]. If the total flow of shared bicycles S(t) is relative to the maximum capacity Cbike of the road segment, then: ； When the duration of this condition exceeds the minimum duration threshold of the peak period for a single vehicle, a "single vehicle tidal return" status label is generated. If the first derivative of the logistics vehicle flow rate satisfies: ； If the corresponding vehicles are mainly concentrated at adjacent intersections less than 2km apart, a "static intersection congestion" status label will be generated. S29. Organize the semantic labels of the status labels within each time interval that meets the conditions into a structured set L of traffic status. , to The semantic label represents the first to the nth semantic label that meets the conditions; the structured semantic label set L contains multiple semantic labels, each label is accompanied by a set of standard state feature vectors generated by summarizing historical multimodal traffic data, which are used to characterize the standard state features of the semantic label in multiple dimensions such as Bluetooth user changes, single vehicle tidal degree, logistics vehicle fluctuation and mixed traffic degree. Each semantic tag includes: tag type, time interval, corresponding intersection code or geographic location information, and feature variable values ​​supporting tag judgment; each semantic tag in the traffic state structured set L matches one or more signal timing templates. This establishes a mapping relationship and constructs a corresponding signal timing template library T. , to This represents the signal timing template from the first signal timing template to the nth signal timing template that meets the conditions for each semantic tag; Signal timing template This includes parameters such as green light duration, red light delay, phase switching sequence, and priority parameters for specific lanes.

5. The intelligent traffic control method for cities according to claim 1, characterized in that, The specific steps for deploying and calling the template library in step two are as follows: The constructed traffic state structure set L and the corresponding signal timing template library T are encapsulated into a unified data structure to generate the first control index table, specifically in SON or XML format, which facilitates network transmission and embedded decoding. The data is pushed to the central control platform via an edge router using serial bus RS485 or CAN, network protocols IP / TCP, MQTT, or WebSocket. The centralized control platform reads the first control index table and extracts the k-th state segment of each sampling window. If the tag condition is met, and tag L is matched, immediately invoke the bound signal timing template. ,include: Change the green light duration, start the left turn phase in advance, give priority to non-motorized vehicles, and coordinate the start of traffic lights.

6. The intelligent traffic control method for cities according to claim 1, characterized in that, The specific steps of real-time multimodal overlap fusion recognition in step three are as follows: S31. A multimodal dataset obtained from edge nodes or sensing devices, including Bluetooth user count N(t), shared bicycle entry and exit volume S(t), video recognition pedestrian density V(t), and logistics vehicle flow Q(t), is executed as follows: Time alignment processing ensures that data from each channel corresponds to the same state segment at the same time t. Feature normalization processing normalizes various types of data to a unified interval [0,1]. Projecting onto a unified feature space generates a fused feature vector: S32. Perform cosine similarity matching between each fused feature vector and the standard state features in the traffic state structure set L to find the matching state label, and call the corresponding matching signal timing template T. i ; If multiple tags simultaneously meet the matching conditions, they are recorded as a multi-tag state set. Real-time collection and fusion of multi-source traffic data to form the current time slice Feature vector Calculate cosine similarity : ； If there exists an 's'-th element that satisfies the condition, then label it. satisfy If the value is ≥0.85, it is considered a "successful match", and then the matching signal timing template T is called accordingly. i .

7. A smart traffic control method for cities according to claim 6, characterized in that, The specific steps of real-time multimodal overlap fusion recognition in step three also include: S33. If the same intersection is in the current time slice When matching multiple state labels, the weight corresponding to the j-th semantic label is used. ∈[0,1], calculate the multi-mode overlap index : ； in, This represents the total number of semantic tags that were successfully matched. Indicates the first The weight of each semantic label; If the multi-mode overlap index A value greater than 0.6 indicates multiple overlapping states, triggering the first adjustment strategy. This strategy prioritizes the use of a "multi-mode collaborative template," extending the green light duration for pedestrian lights, non-motorized vehicle lights, and freight lights by 10%–15%, 8%–12%, and 5%, respectively. It also advances the left-turn phase by 5–7 seconds to reduce intersection waiting conflicts. Non-motorized vehicles are given priority as first-level right-of-way, ensuring their passage during peak pedestrian and bicycle traffic periods. Furthermore, a coordination strategy with adjacent intersection traffic lights is implemented to improve tidal flow traffic by at least 18%. If the multi-mode overlap index A value of <0.3 indicates that the current traffic condition is clearly defined and has a single characteristic, and the previous control period should be maintained without needing to be advanced.

8. A smart traffic control method for cities according to claim 7, characterized in that, Step four includes: S41. If the same intersection is in the current time slice Within the area, there is a simultaneous increase in logistics vehicle traffic and pedestrian / non-motorized vehicle traffic; therefore, the logistics and commuting interference index is calculated. : ； in, This indicates the capacity of the mixed flow channel, which is preset based on the intersection's geographical structure, number of lanes, and width of pedestrian and non-motorized vehicle lanes. , and Each is a current time slice Bluetooth user count, bicycle entry and exit volume, and logistics vehicle traffic; S42, if A value of ≥1.2 indicates that the flow of logistics vehicles has posed a first risk of interference to commuter / non-motorized vehicle traffic, generating a second adjustment strategy, including: delaying the release phase of logistics vehicles, reducing the release time per cycle by 20% to 30%; designating 30% of the original freight lanes as "time-limited slow-moving lanes" or "temporary U-turn lanes" for a duration of no less than 15 minutes; increasing the green light duration by 12% to 20%; and responding to pedestrian crossing signals 3 seconds earlier. If the interference continues for more than 3 cycles, with each cycle lasting 5 minutes, then a structural traffic restriction mode will be implemented, requiring logistics vehicles at the intersection to detour via the auxiliary road. If 0.8≤ ≤1.2 indicates a second risk of interference between pedestrian and vehicle traffic, generating a third adjustment strategy, including: adjusting the release delay of logistics vehicles by 10% to 15%; extending the green light switching buffer by 3 to 5 seconds to prevent overlapping conflicts between pedestrians and vehicles; releasing shared bicycles 5 seconds earlier with the green light; and triggering pop-up guidance to avoid logistics channels in densely populated shared bicycle sections. like A value of <0.8 indicates that there is no risk of mutual interference between logistics and commuting flows, and no special intervention is required; the previous adjustment strategy should continue to be implemented.

9. A smart traffic control method for cities according to claim 8, characterized in that, Step four also includes: S43. After the first, second, or third adjustment strategy is issued and executed, collect the ratio of the current green light pedestrian and vehicle flow at the intersection to the previous cycle, and calculate the traffic efficiency improvement value. : ； ； in, Indicates total traffic flow; like This indicates that traffic efficiency has improved, and we proceed to the next step to determine whether the strategy has achieved its optimization goal. like If the current strategy fails to perform satisfactorily, an emergency rollback mechanism will be implemented to restore the previous time slice. The throughput corresponding to −1) The corresponding strategy for -1); S44. When setting the target passage policy improvement value Te for the current policy, calculate the current policy deviation value ‖E‖: ； A preset error tolerance Eth is set. If ||E|| exceeds Eth, it is determined that the deviation has not converged, and the next round of strategy iteration is initiated. If ||E|| does not exceed Eth, the deviation is considered to have converged, thus achieving the optimization objective, and no adjustment is needed.

10. A smart traffic control system for cities, applied to the smart traffic control method for cities as described in any one of claims 1-9, characterized in that, include: The multi-source data acquisition module is used to deploy Bluetooth detectors, video cameras, non-motorized vehicle detection coils and logistics lane flow meters at key intersections in multi-modal traffic areas to collect real-time data on Bluetooth passenger numbers N(t), single vehicle entry and exit volume S(t), and logistics vehicle flow Q(t) and their geographical location information. Based on the detected network bandwidth and signal stability, the sampling frequency is adaptively adjusted to generate a multimodal dataset; The adaptive sampling processing module periodically detects and records the data distortion status indicators of each channel, including Bluetooth recognition jitter, video image blur, logistics vehicle detection packet loss rate and bandwidth utilization. It constructs and evaluates the multimodal data comprehensive distortion index S, and generates the first throttling sampling strategy or the second sampling strategy. The traffic semantic label modeling module is used to construct a structured set of traffic states L and a corresponding signal timing template library T based on the bus-subway-shared bicycle transfer scenario; and to send L and T to the controllers of each intersection. The overlapping state matching module is used to fuse multimodal datasets obtained from edge nodes or sensing devices, including Bluetooth user numbers N(t), shared bicycle entry and exit volume S(t), video recognition pedestrian density V(t), and logistics vehicle flow Q(t), to generate fused feature vectors. Each fused feature vector is then matched with the standard state features in the traffic state structured set L using cosine similarity to find the matching state label, and the corresponding matching signal timing template T is called. i If the same intersection is in the current time slice When matching multiple state labels, the weight corresponding to the j-th semantic label is used. Calculate the multi-mode overlap index If multiple states overlap, generate the first adjustment strategy; Interference assessment module, if the same intersection is in the current time slice Within the area, there is a simultaneous increase in logistics vehicle traffic and pedestrian / non-motorized vehicle traffic; therefore, the logistics and commuting interference index is calculated. And if there is a risk of interference, a second or third adjustment strategy will be generated. The strategy evaluation and closed-loop optimization module, after the first, second, or third adjustment strategy is issued and executed, collects the ratio of the current green light pedestrian and vehicle flow at the intersection to the previous cycle, and calculates the improvement value of traffic efficiency. The deviation value ‖E‖ from the current strategy is set, and an error tolerance Eth is preset. If the deviation ‖E‖ exceeds Eth, parameter self-learning is triggered, L and T are updated, and iterative optimization is achieved until ‖E‖ converges to the allowable range.

Images