Intelligent traffic light control system and control method thereof

Abstract

The invention discloses an intelligent traffic light control system and a control method thereof. The intelligent traffic light control system comprises a sensing layer which integrates a multi-mode sensor to collect traffic data in real time, and an edge computing node completes data preprocessing and dynamically adjusts signal light parameters; the transmission layer is used for performing high-speed data transmission and emergency request security verification through 5G / C-V2X, low-power-consumption Bluetooth and a block chain network, and adopting AES-256 encryption to guarantee data security; the platform layer is used for sharing an AI model based on a federated learning framework to improve prediction accuracy, optimizing phase switching by combining a dynamic timing engine with flow trend analysis and reinforcement learning, and triggering emergency vehicles to pass preferentially by a block chain verification module; the application layer is used for carrying out remote control and user feedback through a mobile phone APP, assisting in displaying a signal state and recommending a vehicle speed through AR navigation, managing a background monitoring equipment state and supporting firmware upgrading; and efficient traffic management in a temporary scene is carried out through a full-link closed-loop design.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traffic light control, and in particular to an intelligent traffic light control system and a control method thereof. BACKGROUND

[0002] A temporary traffic signal light is a portable device for temporary traffic control, which is usually used to replace or assist fixed signal lights in road construction, traffic accident handling or large-scale activities, and controls the traffic order of vehicles and pedestrians through the alternating display of red, yellow and green lights, so as to ensure safety and smoothness in temporary traffic scenes, and has the characteristics of convenient installation, flexible movement and strong adaptability.

[0003] In the scenes of construction sections, school periphery and large-scale activity sites where temporary traffic lights are widely used, the traditional temporary signal light often faces the dilemma of insufficient dynamic adjustment capability due to the dependence on fixed timing scheme. For example, during the school dismissal period of 16:30-17:30 every day, the demand for pedestrian crossing increases sharply, but the traditional temporary light still follows the flat peak period timing, which leads to the phenomenon of pedestrian retention at the intersection and vehicle long-time waiting, and even causes the phenomenon of red light running. Therefore, an intelligent traffic light control system and a control method thereof are proposed. SUMMARY

[0004] The purpose of the present application is to solve the problems existing in the prior art and provide an intelligent traffic light control system and a control method thereof.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme: An intelligent traffic light control system comprises: A perception layer: traffic data is collected in real time through a multi-modal sensor array (camera, millimeter wave radar, infrared sensor, environmental sensor), the camera identifies vehicle flow, pedestrian and non-motor vehicle behavior, the millimeter wave radar detects vehicle speed and distance, the infrared sensor covers the low-light scene at night, and the environmental sensor monitors weather parameters; an edge computing node pre-processes the original data, such as target detection, classification and denoising, and dynamically adjusts the signal light brightness and mode based on the environmental data, such as enhancing the visibility in rainy and foggy weather; the processed structured data (vehicle flow density, pedestrian quantity, abnormal event) is uploaded to the platform layer through a transmission layer, and at the same time, key data is locally reserved for rapid decision-making, such as emergency vehicle priority triggering; Transmission layer: Build a multi-channel communication network, 5G / C-V2X module transmits perception layer data to platform layer at high speed, supports vehicle-road cooperation such as broadcasting signal lamp status to nearby vehicles; Bluetooth low energy performs short-distance interaction between mobile phone APP and signal lamp, such as adjusting signal period or triggering mobile; Blockchain network encrypts and transmits emergency vehicle request and verifies it; Data is managed by priority queue (real-time video stream > emergency instruction > general data) to reduce delay, while AES-256 encryption is used to ensure data security and prevent tampering or leakage; Platform layer: Integrate multi-source data and execute intelligent decision-making, federal learning framework shares AI model across intersections, combines local training data to improve global prediction accuracy; Dynamic timing engine predicts short-term traffic trends based on ARIMA algorithm, analyzes current traffic in real time through gated Transformer, and optimizes phase switching through reinforcement learning to minimize total delay time; Blockchain verification module receives encrypted requests from emergency vehicles, verifies identity through smart contract, triggers signal lamp to switch to priority phase, and extends green light time. The final control instructions such as phase adjustment and green light extension are issued to the application layer through the transmission layer; Application layer: Mobile phone APP receives platform layer instructions, supports user remote signal lamp movement, signal time adjustment or temporary regulation mode setting; Real-time display of signal lamp status, estimated waiting time and emergency vehicle warning, and integration of user feedback entry to generate congestion report for dynamic optimization of timing; AR navigation assistance shows signal lamp countdown, recommended speed and abnormal event prompt to drivers through vehicle-mounted devices; Management background monitors all signal lamp states (position, power, fault), supports historical data backtracking and remote firmware upgrade to ensure stable system operation.

[0006] The above technical solution further includes: Further, the edge computing node pre-processes the raw data and dynamically adjusts the signal lamp brightness and mode based on environmental data, including the following steps: The edge computing node first receives raw data from a multi-modal sensor array: Camera: Real-time video stream, including vehicle flow, pedestrian, non-motor vehicle behavior; Millimeter wave radar: Vehicle speed, distance, lane occupancy rate, and other structured data; Infrared sensor: Thermal imaging data in low light night scene (identify pedestrians, animals, etc.); Environmental sensor: Temperature, humidity, visibility, light intensity, and other environmental parameters; All data is synchronized by timestamp to ensure that multi-source information is aligned in time dimension, providing a basis for subsequent fusion analysis; The edge computing node performs lightweight processing on raw data to reduce redundant data transmission to the platform layer; removes noise in camera video streams, such as rain and fog interference, lens stains causing blurring; corrects abnormal distance measurement values of millimeter wave radar due to multipath effects; filters false detection signals of infrared sensors due to environmental heat radiation, such as direct sunlight; camera data: identifies vehicle types (sedans, trucks), pedestrian numbers and behaviors (running red lights, staying) through lightweight target detection algorithms; calculates lane-level traffic density and average speed; marks the thermal imaging profile of pedestrians / animals and determines their moving direction (crossing the road, walking along the road); quantifies visibility and light intensity, and classifies weather conditions (sunny / rainy / foggy / nighttime); Based on the pre-processed environmental parameters, the edge computing node dynamically evaluates the signal light visibility requirement; Low-light scenes (such as night, tunnel): When the infrared sensor detects pedestrians or animals, the signal light is triggered to switch to "night mode" (enhance brightness to more than 5000 nits, use high-contrast red-yellow-green color scheme); when the light intensity sensor detects that the light intensity is less than 100 lux, the pedestrian green light time is automatically extended to ensure pedestrian safety; Severe weather scenes (such as rain and fog): When the visibility sensor detects that the visibility is less than 50 meters, the signal light switches to "fog mode" (brightness is increased to 3000 nits, and flashing warning function is used); when the millimeter wave radar detects that the lane-level traffic density decreases, the signal cycle is temporarily shortened to reduce red light waiting time and avoid congestion caused by low visibility; The edge computing node directly triggers signal light adjustment instructions based on environmental analysis results without waiting for platform layer response; dynamically adjusts the current of LED beads through PWM pulse width modulation technology for stepless brightness adjustment; calls preset signal light display templates such as night mode and fog mode, which include brightness, color contrast, and flashing frequency parameters; in extreme scenarios, such as sudden accidents causing complete lane blockage, the edge node can temporarily extend the current phase green light time (up to 30 seconds) and simultaneously report abnormal events to the platform layer to request subsequent global timing optimization; The edge computing node uploads pre-processed structured data (such as traffic density, pedestrian number, and environmental parameters) and local adjustment records (brightness change, mode switching) to the platform layer through the transmission layer; optimizes the mapping relationship between environmental parameters and signal light adjustment strategies, such as "visibility 30 meters → brightness 4000 nits"; the platform layer verifies the effectiveness of the global algorithm in extreme scenarios based on the local adjustment records of the edge node and dynamically corrects timing parameters, such as adjusting the signal cycle reference value in rainy and foggy weather.

[0007] Furthermore, the blockchain network encrypts and verifies emergency vehicle requests; data is managed through priority queues to reduce latency, and AES-256 encryption is used to ensure data security, including the following steps: Emergency vehicles generate encrypted requests via onboard terminals or dedicated equipment, including vehicle identification information (such as license plate number and rescue type), current location, target route, and priority. The request content is sent to the nearest temporary traffic light transmission layer node via the onboard 5G / C-V2X module. After receiving the request, the transmission layer node packages the raw data into a blockchain transaction, uses an asymmetric encryption algorithm to generate a digital signature, and ensures that the request is tamper-proof. The encrypted transaction is broadcast to all participating nodes, such as other temporary traffic lights and traffic management centers, through the blockchain network, forming a distributed ledger record. Blockchain nodes run smart contracts for automatic verification: Vehicle identity: The vehicle's legality is confirmed by comparing it with a pre-defined whitelist (such as the license plates of ambulances registered with emergency centers); Request timeliness: Check if the request was generated within the validity period; Priority matching: Set priority thresholds based on the type of rescue (such as fire, medical emergency) to ensure that high-priority requests are processed first; Once the verification is successful, the smart contract generates a "priority pass" and triggers a traffic light adjustment command; The transport layer node sends the verified "priority passage credential" to the platform layer via the 5G / C-V2X module. The platform layer's dynamic timing engine immediately adjusts the traffic light phase (e.g., extending the current green light time by 30 seconds, or forcibly switching to the emergency vehicle passage direction) and simultaneously broadcasts the traffic light change to nearby vehicles, such as "An ambulance is about to pass, please do not run the red light." After the traffic light adjustment is executed, the execution result, such as "The green light has been extended to 45 seconds," is sent back to the blockchain network via the transport layer, forming a closed-loop verification. The transport layer classifies and prioritizes all uplink (sensor layer → platform layer) and downlink (platform layer → application layer) data. High priority: emergency vehicle requests, real-time video streams (traffic accident footage captured by cameras), traffic light adjustment instructions; Medium priority: Structured sensor data (vehicle density, number of pedestrians), user feedback (congestion reports); Low priority: Historical data backhaul (daily traffic statistics), system logs.

[0008] The transport layer employs a "strict priority queue" algorithm: High-priority data uses dedicated bandwidth, such as the emergency channel reserved in the 5G network, to ensure end-to-end latency <100ms; Medium-priority data is allocated resources in the remaining bandwidth according to a "weighted fair queue" to avoid low-priority data consuming too much bandwidth; Low-priority data is transmitted only when bandwidth is idle, and dynamic discarding is supported (e.g., automatically discarding historical logs when the network is congested). The transport layer monitors network load in real time, such as bandwidth utilization and queue length. When congestion is detected, high-priority data remains unchanged, medium-priority data reduces its transmission rate, and low-priority data is suspended from transmission until the network recovers. All uplink data (perception layer → platform layer) and downlink data (platform layer → application layer) are encrypted using AES-256 before transmission. Perception layer data (such as camera video streams and sensor values) are encrypted at the edge computing node; platform layer commands (such as traffic light adjustment parameters) are encrypted in the management backend; and user feedback data (such as congestion reports submitted by the APP) are encrypted on the mobile APP.

[0009] Key management adopts a "hierarchical key system". The master key is generated uniformly by the traffic management center and replaced periodically. The session key is dynamically generated by the transport layer node, used for a single communication, and is automatically destroyed after expiration. The encrypted data is transmitted to the receiving end through the transport layer network (5G / C-V2X, Bluetooth Low Energy). The receiving end (such as the platform layer server or mobile APP) uses the corresponding session key to decrypt the data and restore the original content. If decryption fails, such as due to key mismatch or data tampering, the receiving end immediately discards the data and triggers an alarm, such as reporting an "encryption anomaly event" to the management backend. All encrypted data transmissions are recorded with hash values ​​through the blockchain network, forming an immutable audit log. The management backend can query historical data transmission records at any time to verify data integrity, such as confirming that an emergency vehicle request has not been tampered with. If a security vulnerability is discovered (such as key leakage), the data flow can be traced through the blockchain to quickly locate and isolate the affected nodes.

[0010] Furthermore, the federated learning framework shares AI models across intersections and combines them with local training data to improve global prediction accuracy, including the following steps: Each temporary traffic light (intersection node) is loaded with a pre-trained basic AI model during deployment. This model has been initialized using public traffic datasets (such as urban road traffic flow and pedestrian behavior patterns) and has basic traffic prediction capabilities. The intersection node uses multimodal data (vehicle density, pedestrian numbers, emergency vehicle request records, etc.) uploaded from the perception layer, combined with local environmental parameters, such as school dismissal times and peak hours in commercial areas, to generate a labeled dataset. Input features: historical traffic density over the past 5 minutes, current weather conditions, and traffic light status at nearby intersections; Output objectives: Traffic flow forecast for each lane and pedestrian crossing demand probability for the next 10 minutes; The intersection nodes run local training tasks on edge computing devices and fine-tune the base model using labeled datasets, such as adjusting the attention mechanism weights of the gated Transformer to adapt the model to the unique traffic patterns of the intersection, such as the high proportion of pedestrians at intersections near schools during morning and evening rush hours and the high volume of trucks at intersections in industrial areas. During training, only the model parameters (weight matrix, bias terms) are retained, and the original data is always stored locally and not transmitted to external systems. After local training is completed, the intersection nodes encrypt and upload the updated model parameters (rather than the original data) to the central federated learning server (deployed in the traffic management center). Differential privacy techniques (such as adding Gaussian noise) are used during upload to further protect local data privacy (avoiding the inference of traffic flow values ​​from parameters). After receiving the model parameters from all intersection nodes, the federated server runs an aggregation algorithm. Weight calculation: Weights are assigned based on the amount of data at intersection nodes (such as the number of traffic density samples), with nodes having a larger amount of data contributing more; Anomaly detection: By statistically calculating the Euclidean distance of the weight matrix of each node through parameter distribution, abnormal nodes are identified and eliminated, such as those whose parameters deviate from the normal range due to sensor failure. The aggregated global model parameters generate a new "global baseline model," which integrates traffic pattern characteristics of multiple intersections, such as traffic flow patterns under different functional zones and weather conditions. The federated server distributes the aggregated global model parameters to all intersection nodes, overwriting the original local models. After receiving the global model, the intersection nodes combine it with the latest data they have stored locally and perform secondary fine-tuning, such as adjusting the output layer of the global model to match the traffic light phase configuration of the intersection, ensuring that the model has both global prediction capabilities and can adapt to local details. The updated model is used at intersection nodes to predict traffic flow and pedestrian demand in real time for the next 10 minutes (e.g., "Traffic flow from east to west on the main road will increase by 30% in the next 5 minutes"). The prediction results are then input into the dynamic timing engine to adjust the traffic light phases and extend the green light time for east to west. The prediction results are also sent back to the federated server as a reference for the next round of global model aggregation (intersections marked with "higher prediction errors during school dismissal time" will have their weight increased in subsequent aggregations). The federal server sets fixed periods or trigger conditions (detection of major traffic events or sudden weather changes) to initiate a new round of model aggregation and distribution, ensuring that the global model always reflects the latest traffic patterns (changes in traffic flow during holidays and the impact of newly opened roads). The raw data is always stored locally. The federated server only has access to the encrypted model parameters and cannot infer the traffic flow details at specific intersections. For example, it cannot determine from the parameters that "there are 100 cars at a certain intersection at 7 a.m.". The variance of the parameters of each node is calculated by monitoring the parameter distribution. If the parameters of a node are detected to deviate continuously from the global mean (due to sensor failure or malicious attack), its model upload permission is immediately suspended and it is marked as requiring manual inspection.

[0011] Furthermore, the dynamic timing engine predicts short-term traffic trends based on the ARIMA algorithm and analyzes the current traffic flow in real time through a gated Transformer, including the following steps: Input historical traffic flow data (traffic density every 5 minutes in the past hour), eliminate non-stationarity (trend, seasonality) through difference operation, determine the difference order d; calculate the autocorrelation function and partial autocorrelation function, determine the autoregression order p and the moving average order q, and finally select the ARIMA model as ARIMA(p,d,q); Use maximum likelihood estimation to fit the model parameters (autoregressive coefficients). , , ..., Moving average coefficient , , ..., (Constant term c); Based on the fitted model, predict the traffic flow trend in the next h steps and output the predicted value. ; Calculate prediction error ( (Based on actual traffic flow), if the error exceeds the threshold (e.g., mean square error MSE > 10%), the model order is redefined and the parameters are adjusted. Core formula:

[0012] in, Let be the predicted traffic density at time t; The actual traffic density at time ti; The prediction error at time tj; , represents the autoregressive and moving average coefficients; c is a constant term (such as average traffic density). Input real-time traffic density data ( ), number of pedestrians ( ), environmental parameters (such as visibility) Weather conditions Data is mapped to vectors through a linear embedding layer. , dimension ; Calculate the query, key, and value matrix: , ,

[0013] in, , , The learnable parameter matrix; Calculate the attention weight A (scaled dot product attention):

[0014] in, The key vector dimension is used for scaling to prevent gradient vanishing; the attention output and the original input are fused through a gating mechanism.

[0015] Where σ is the Sigmoid function, represents the gating parameter, and ⊙ represents element-wise multiplication; The gated Transformer outputs a comprehensive feature vector of the current traffic flow. The data is input to the fully connected layer to generate real-time traffic flow analysis results (e.g., "The current traffic density on the main road from east to west is 20 vehicles / minute, and the probability of pedestrian crossing demand is 30%").

[0016] Where Q, K, and V are the query, key, and value matrices, respectively; A is the key vector dimension (used for scaling); A is the attention weight matrix (representing the importance of each time step / feature); V is the value matrix (containing original data information); ARIMA provides traffic flow trends for the next 10 minutes, while the gating Transformer analyzes the current traffic flow status. The results from both are fed into the dynamic timing engine. Based on the fluctuations in real-time traffic flow, such as a sudden event causing a surge in traffic flow, the output weight of the gating Transformer will be dynamically increased to ensure that traffic light adjustments both reference long-term trends and respond to immediate changes.

[0017] Furthermore, the blockchain verification module receives an encrypted request from an emergency vehicle, verifies its identity via a smart contract, and then triggers a forced switch of the traffic lights to the priority phase, including the following steps: Emergency vehicles (ambulances, fire trucks) generate encrypted requests via onboard terminals or dedicated equipment, containing the following information: Identification information: Unique vehicle identifier (such as license plate number, rescue unit registration number); Location and Route: Current GPS coordinates, target location (e.g., hospital, fire scene), planned route (latitude and longitude sequence); Priority: The priority level corresponding to the type of rescue (e.g., "Level 1 Fire" or "Level 2 Medical Emergency"); Timestamp: The specific time the request was generated; The vehicle terminal sends the encryption request to the nearest temporary traffic light transmission layer node via the 5G / C-V2X communication module, enabling low-latency transmission; After receiving the request, the transport layer node encapsulates it into a blockchain transaction, which includes: Raw data: Plaintext information of emergency vehicle requests (identity, location, priority, etc.); Digital signature: A signature is generated using an asymmetric encryption algorithm to ensure that the transaction cannot be tampered with; Hash value: Calculate the SHA-256 hash of the original data, which serves as a unique identifier for the transaction; The encapsulated transaction is broadcast to all participating nodes (including other temporary traffic lights and traffic management center servers) via the blockchain network, forming a distributed ledger record. Each node independently stores a copy of the transaction, ensuring data redundancy and resistance to attacks. Blockchain nodes run pre-defined smart contracts (written in Solidity) to automatically perform verification: Identity legitimacy: By comparing with the whitelist pre-stored by the traffic management center (such as the license plate of an ambulance registered with the emergency center), it is confirmed whether the vehicle is a legitimate rescue unit; Request timeliness: Check if the timestamp is within the validity period (e.g., the difference between the request generation time and the current time is less than 5 minutes) to prevent expired or forged requests; Priority matching: Set priority thresholds based on the type of rescue (e.g., "Level 1 Fire" > "Level 2 Medical Emergency") to ensure that high-priority requests are processed first; Upon successful verification, the smart contract generates a "priority pass credential" (containing a hash value, verification result, and validity period) and triggers a traffic light adjustment command. If verification fails, such as when the vehicle is not on the whitelist or the request has expired, the smart contract directly rejects the request and records an alarm log. The transport layer node sends the "priority pass credential" generated by the smart contract to the platform layer dynamic timing engine via the 5G / C-V2X module, and the engine executes it immediately: Forced phase switching: Based on the planned route of emergency vehicles (such as "traveling east to west along the main road"), switch the traffic light in the target direction to green and extend the green light time; Conflict direction suppression: Temporarily switch the light to red for directions that conflict with emergency vehicle routes to avoid intersection congestion; The platform layer sends broadcast messages to vehicles near the emergency vehicle's path via the 5G / C-V2X module: traffic light status (e.g., "Eastbound green light has been extended to 60 seconds"), recommended speed, and abnormal warnings (e.g., "Ambulance is about to pass, please do not run the red light"). After the traffic light adjustment is completed, the transmission layer nodes package the execution results (such as "green light extended to 60 seconds" and "no conflicting vehicles") into blockchain transactions and send them back to the blockchain network, forming a closed-loop verification. All nodes update their ledger records to ensure data consistency. When requesting packaging, random noise is added to non-critical data (such as the specific vehicle model) to prevent the inference of detailed vehicle information from the blockchain data; if a node continues to send invalid requests (such as forging whitelisted vehicles), the blockchain network marks it as abnormal through the consensus mechanism and suspends its transaction processing authority until manual intervention is required for verification.

[0018] Furthermore, the integrated user feedback entry point, after verification, dynamically optimizes the timing, including the following steps: Users can access the "Traffic Feedback" module through the mobile app, select the feedback type (e.g., "Congestion," "Accident," "Traffic Light Malfunction"), and fill in the required information. Location: Automatically locate the current position or manually select the intersection; Time: Defaults to the current time, but can be manually adjusted (e.g., to report past traffic congestion). Description: Text input box (e.g., "Vehicles traveling from east to west are queuing up to 50 meters outside the intersection"). Evidence: You can choose to upload images (such as camera screenshots), videos (such as dashcam footage), or voice descriptions. The app performs initial screening of feedback, filtering out invalid content (such as blank descriptions or lack of location information) and automatically marking feedback as urgent, such as when the user selects the "Accident" type and marks it as "Requires Urgent Handling"; Feedback data is sent to the platform layer via the 5G / C-V2X module or Bluetooth Low Energy in the transport layer, and is encrypted with AES-256 to ensure that the data is not stolen or tampered with during transmission; emergency feedback is marked as high priority and transmitted through dedicated bandwidth; ordinary feedback (such as congestion) is marked as medium priority. After receiving feedback, the platform layer verifies it by combining data from the perception layer, historical records, and other user feedback: Comparison with sensor data: If a user reports "congestion in the eastbound direction", the system checks the camera and radar data at the intersection to confirm whether the traffic density exceeds the threshold (e.g., more than 20 vehicles / minute). Cross-validation with other user feedback: If multiple users report the same problem in the same location and at similar times, the credibility is increased; Historical behavior analysis: If a user has submitted false feedback multiple times in the past (such as false reporting of congestion), the system reduces the weight of their feedback; If the feedback involves an emergency event, the platform layer can call a blockchain smart contract to automatically verify the user's identity (such as whether they are a registered driver) and the reasonableness of the feedback content (such as whether the accident description conforms to common patterns); The verified feedback is input into the dynamic timing engine, and the engine executes according to the feedback type: Congestion feedback: Extend the green light time in the congested direction, or adjust the signal phases of adjacent intersections to divert vehicles; Accident feedback: Temporarily switch the signal lights at the accident intersection to the "accident mode", such as extending the red light time in all directions to prioritize the passage of rescue vehicles; Signal light anomaly feedback: If a user reports that "the green light in a certain direction is not on", the system remotely diagnoses the hardware status of the signal light. If a fault is confirmed, it switches to the backup signal light mode, such as enabling temporary indicator lights; Frequent congestion feedback data (such as a certain intersection being congested from 17:00 to 18:00 every day) is incorporated into the federated learning framework as local training data to update the AI model at that intersection, improving the accuracy of subsequent traffic flow predictions, such as the model learning to "increase the green light time from east to west after 17:00"; The platform layer pushes processing notifications to users through the mobile APP: Feedback status: "Verified and processed" or "Failed verification (evidence needs to be supplemented)"; Details of timing adjustment: Such as "The green light from east to west has been extended to 45 seconds, expected to relieve congestion"; Reward mechanism: Users can obtain points for each valid feedback, and the points can be used to redeem benefits such as parking coupons, incentivizing continuous participation.

[0019] The adjusted signal light timing parameters are sent to the edge computing nodes through the transport layer. The nodes update the local signal light control logic and synchronize it to the cameras and radars in the perception layer to ensure that the sensor data is consistent with the timing status. For example, the camera adjusts the pedestrian detection area according to the new green light time; If a user frequently submits false feedback (such as false reporting of congestion), the system marks them as an "abnormal user" through behavior analysis (such as the feedback location not matching the actual GPS) and restricts their feedback permissions; Personal information involved in the user's feedback, such as the license plate number, is desensitized, such as replaced with "Yue B****", only retaining the necessary location and event description.

[0020] A control method for an intelligent traffic light control system includes: Step 1: Real-time traffic data is collected by a multimodal sensor array (cameras, millimeter-wave radar, infrared sensors, and environmental sensors) deployed at the intersection, covering traffic density, pedestrian behavior, vehicle speed and distance, weather parameters, etc.; edge computing nodes clean the raw data (remove noise), extract features (identify the number of pedestrians and the direction of traffic flow), and adapt to the environment (such as enhancing the sensitivity of infrared sensors in low light conditions at night), generating structured data that is uploaded to the transmission layer, while key data (such as the location of emergency vehicles) is retained locally for rapid decision-making; Step 2: The 5G / C-V2X module transmits the structured data from the perception layer to the platform layer at high speed, supporting vehicle-road cooperation; Bluetooth Low Energy enables short-range interaction between the mobile app and traffic lights, such as users adjusting signal timing; the blockchain network encrypts and transmits emergency vehicle requests (such as ambulances) and verifies smart contracts (verifying whitelists and timeliness), and the data is managed through priority queues (high-priority data occupies dedicated bandwidth) to reduce latency, and AES-256 encryption is used to ensure end-to-end security; Step 3: Integrate multi-source data and share AI models across intersections using a federated learning framework (combining local training data to improve global prediction accuracy); the dynamic timing engine predicts short-term traffic trends based on the ARIMA algorithm, analyzes the current traffic flow in real time through a gating Transformer, and then optimizes phase switching (extending green light time) through reinforcement learning to minimize total delay time; the blockchain verification module receives encrypted requests, verifies identity through a smart contract, and then triggers a forced switch of the traffic lights to the priority phase, for example, extending the green light for ambulances to 60 seconds; Step 4: Control commands generated at the platform layer, such as phase switching and brightness adjustment, are sent down to the application layer for execution via the transmission layer. The traffic light hardware directly adjusts the phase and brightness. The mobile app displays the traffic light status, estimated waiting time, and emergency vehicle warnings in real time. AR navigation assistance displays a countdown and recommended speed through in-vehicle devices, ensuring precise execution of control commands and improving user interaction efficiency. Step 5: Integrate user feedback entry points (reporting congestion, accidents) through the mobile APP, and dynamically optimize timing (adjusting the weights of the federated learning model) after cross-validating the feedback content (comparing it with sensor data); frequently reported data (such as daily congestion at a certain intersection at 17:00) are incorporated into the federated learning framework to update the local AI model and improve the accuracy of subsequent predictions; the management backend monitors the status of traffic lights, supports historical data backtracking and remote firmware upgrades, forming a closed loop of "execution-feedback-optimization".

[0021] The present invention has the following beneficial effects: In this invention, traffic data is collected in real time through a multimodal sensor array, preprocessed by edge computing nodes, and then uploaded to the platform layer. A multi-channel communication network is constructed to ensure high-speed and secure data transmission. Multi-source data is integrated, and a dynamic timing engine is built using a federated learning framework, ARIMA algorithm, gated Transformer, and reinforcement learning. Combined with a blockchain verification module, emergency vehicle requests are processed to achieve traffic light phase switching and timing optimization. Furthermore, a mobile APP, AR navigation, and management backend support remote user interaction, real-time status display, and feedback optimization, effectively solving the problem of insufficient dynamic adjustment capability of traffic lights. Attached Figure Description

[0022] Fig. 1 This is a system block diagram of an intelligent traffic light control system proposed in this invention; Fig. 2 This is a flowchart illustrating the steps of a control method for an intelligent traffic light control system proposed in this invention. Detailed Implementation

[0023] 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.

[0024] Please see Figs. 1-2 As shown, the present invention is an intelligent traffic light control system and its control method, comprising: An intelligent traffic light control system includes: Perception Layer: Real-time traffic data is collected through a multimodal sensor array (cameras, millimeter-wave radar, infrared sensors, and environmental sensors). Cameras identify vehicle flow, pedestrian and non-motorized vehicle behavior, millimeter-wave radar detects vehicle speed and distance, infrared sensors cover low-light nighttime scenarios, and environmental sensors monitor weather parameters. Edge computing nodes preprocess the raw data, such as target detection, classification, and noise reduction, and dynamically adjust the brightness and mode of traffic lights based on environmental data, such as enhancing visibility in rainy and foggy weather. The processed structured data (vehicle flow density, pedestrian count, and abnormal events) is uploaded to the platform layer through the transmission layer, while key data is retained locally for rapid decision-making, such as prioritizing emergency vehicle triggering. Transport Layer: A multi-channel communication network is constructed. The 5G / C-V2X module transmits perception layer data to the platform layer at high speed, supporting vehicle-to-infrastructure (V2I) communication, such as broadcasting traffic light status to nearby vehicles. Low-power Bluetooth enables short-range interaction between mobile apps and traffic lights, such as adjusting signal cycles or triggering movement. The blockchain network encrypts and verifies emergency vehicle requests. Data is managed through priority queues (real-time video stream > emergency commands > regular data) to reduce latency, while AES-256 encryption is used to ensure data security and prevent tampering or leakage. Platform Layer: Integrates multi-source data and executes intelligent decisions. The federated learning framework shares AI models across intersections and combines local training data to improve global prediction accuracy. The dynamic timing engine predicts short-term traffic trends based on the ARIMA algorithm, analyzes current traffic flow in real time through a gating Transformer, and then optimizes phase switching through reinforcement learning to minimize total delay time. The blockchain verification module receives encrypted requests from emergency vehicles, verifies their identity via a smart contract, and triggers a forced switch of the traffic lights to the priority phase, extending the green light time. The final generated control commands, such as phase adjustment and green light extension, are sent to the application layer for execution through the transport layer. Application Layer: The mobile app receives instructions from the platform layer, allowing users to remotely move traffic lights, adjust signal times, or set temporary control modes; it displays real-time traffic light status, estimated waiting time, and emergency vehicle warnings, and integrates a user feedback entry point to generate congestion reports, dynamically optimizing timings after verification; AR navigation assistance displays traffic light countdowns, recommended speeds, and abnormal event alerts to the driver through in-vehicle devices; the management backend monitors the status of all traffic lights (location, battery level, faults), supports historical data review and remote firmware upgrades, ensuring stable system operation.

[0025] In one embodiment, the edge computing node preprocesses the raw data and dynamically adjusts the brightness and mode of the traffic lights based on environmental data, including the following steps: The edge computing node first receives raw data from the multimodal sensor array: Camera: Real-time video stream, including traffic flow, pedestrians, and non-motorized vehicle behavior; Millimeter-wave radar: structured data such as vehicle speed, distance, and lane occupancy rate; Infrared sensor: Thermal imaging data in low-light nighttime scenarios (for identifying pedestrians, animals, etc.); Environmental sensors: environmental parameters such as temperature, humidity, visibility, and light intensity; All data is synchronized via timestamps to ensure that information from multiple sources is aligned in the time dimension, providing a foundation for subsequent fusion analysis; Edge computing nodes perform lightweight processing on raw data, reducing redundant data transmitted to the platform layer; remove noise from camera video streams, such as rain and fog interference, and blurring caused by lens smudges; correct abnormal distance measurements caused by multipath effects from millimeter-wave radar; filter false detection signals from infrared sensors caused by environmental heat radiation, such as direct sunlight; and use lightweight target detection algorithms to identify vehicle types (cars, trucks), pedestrian numbers and behaviors (running red lights, stopping); calculate lane-level traffic density and average vehicle speed; mark the thermal imaging contours of pedestrians / animals and determine their direction of movement (crossing the road, walking along the road); quantify visibility and light intensity, and classify weather conditions (sunny / rainy / foggy / night). Based on preprocessed environmental parameters, edge computing nodes dynamically assess traffic light visibility requirements; Low-light scenes (such as at night, in tunnels): When the infrared sensor detects a pedestrian or animal, it triggers the traffic light to switch to "night mode" (increased brightness to over 5000 nits, using a high-contrast red, yellow, and green color scheme); when the light intensity sensor detects that the light intensity is less than 100 lux, it automatically extends the green light time for pedestrians to ensure pedestrian safety when crossing the street. Severe weather conditions (such as rain and fog): When the visibility sensor detects visibility of less than 50 meters, the traffic lights switch to "fog mode" (brightness increased to 3000 nits, with flashing warning function); when the millimeter-wave radar detects a decrease in lane-level traffic density, the signal cycle is temporarily shortened to reduce red light waiting time and avoid exacerbating congestion due to low visibility. Based on environmental analysis results, edge computing nodes can directly trigger traffic light adjustment commands without waiting for a platform response; they can dynamically adjust the current of LED beads using PWM pulse width modulation technology to achieve stepless brightness adjustment; they can call preset traffic light display templates such as night mode and fog mode, which include parameters such as brightness, color contrast, and flashing frequency; in extreme scenarios, such as a sudden accident causing a complete lane blockage, edge nodes can temporarily extend the green light time of the current phase (up to 30 seconds) and simultaneously report the abnormal event to the platform layer, requesting subsequent global timing optimization. Edge computing nodes upload pre-processed structured data (such as traffic density, pedestrian numbers, and environmental parameters) and local adjustment records (brightness changes, mode switching) to the platform layer through the transport layer; combined with data from other intersections, they optimize the mapping relationship between environmental parameters and traffic light adjustment strategies, such as "visibility 30 meters → brightness 4000 nits"; the platform layer verifies the effectiveness of the global algorithm in extreme scenarios based on the local adjustment records of the edge nodes, and dynamically corrects timing parameters, such as adjusting the signal cycle reference value in rainy or foggy weather.

[0026] In one embodiment, the blockchain network encrypts and verifies emergency vehicle requests; data is managed through priority queues to reduce latency, and AES-256 encryption is used to ensure data security, including the following steps: Emergency vehicles generate encrypted requests via onboard terminals or dedicated equipment, including vehicle identification information (such as license plate number and rescue type), current location, target route, and priority. The request content is sent to the nearest temporary traffic light transmission layer node via the onboard 5G / C-V2X module. After receiving the request, the transmission layer node packages the raw data into a blockchain transaction, uses an asymmetric encryption algorithm to generate a digital signature, and ensures that the request is tamper-proof. The encrypted transaction is broadcast to all participating nodes, such as other temporary traffic lights and traffic management centers, through the blockchain network, forming a distributed ledger record. Blockchain nodes run smart contracts for automatic verification: Vehicle identity: The vehicle's legality is confirmed by comparing it with a pre-defined whitelist (such as the license plates of ambulances registered with emergency centers); Request timeliness: Check if the request was generated within the validity period; Priority matching: Set priority thresholds based on the type of rescue (such as fire, medical emergency) to ensure that high-priority requests are processed first; Once the verification is successful, the smart contract generates a "priority pass" and triggers a traffic light adjustment command; The transport layer node sends the verified "priority passage credential" to the platform layer via the 5G / C-V2X module. The platform layer's dynamic timing engine immediately adjusts the traffic light phase (e.g., extending the current green light time by 30 seconds, or forcibly switching to the emergency vehicle passage direction) and simultaneously broadcasts the traffic light change to nearby vehicles, such as "An ambulance is about to pass, please do not run the red light." After the traffic light adjustment is executed, the execution result, such as "The green light has been extended to 45 seconds," is sent back to the blockchain network via the transport layer, forming a closed-loop verification. The transport layer classifies and prioritizes all uplink (sensor layer → platform layer) and downlink (platform layer → application layer) data. High priority: emergency vehicle requests, real-time video streams (traffic accident footage captured by cameras), traffic light adjustment instructions; Medium priority: Structured sensor data (vehicle density, number of pedestrians), user feedback (congestion reports); Low priority: Historical data backhaul (daily traffic statistics), system logs.

[0027] The transport layer employs a "strict priority queue" algorithm: High-priority data uses dedicated bandwidth, such as the emergency channel reserved in the 5G network, to ensure end-to-end latency <100ms; Medium-priority data is allocated resources in the remaining bandwidth according to a "weighted fair queue" to avoid low-priority data consuming too much bandwidth; Low-priority data is transmitted only when bandwidth is idle, and dynamic discarding is supported (e.g., automatically discarding historical logs when the network is congested). The transport layer monitors network load in real time, such as bandwidth utilization and queue length. When congestion is detected, high-priority data remains unchanged, medium-priority data reduces its transmission rate, and low-priority data is suspended from transmission until the network recovers. All uplink data (perception layer → platform layer) and downlink data (platform layer → application layer) are encrypted using AES-256 before transmission. Perception layer data (such as camera video streams and sensor values) are encrypted at the edge computing node; platform layer commands (such as traffic light adjustment parameters) are encrypted in the management backend; and user feedback data (such as congestion reports submitted by the APP) are encrypted on the mobile APP.

[0028] Key management adopts a "hierarchical key system". The master key is generated uniformly by the traffic management center and replaced periodically. The session key is dynamically generated by the transport layer node, used for a single communication, and is automatically destroyed after expiration. The encrypted data is transmitted to the receiving end through the transport layer network (5G / C-V2X, Bluetooth Low Energy). The receiving end (such as the platform layer server or mobile APP) uses the corresponding session key to decrypt the data and restore the original content. If decryption fails, such as due to key mismatch or data tampering, the receiving end immediately discards the data and triggers an alarm, such as reporting an "encryption anomaly event" to the management backend. All encrypted data transmissions are recorded with hash values ​​through the blockchain network, forming an immutable audit log. The management backend can query historical data transmission records at any time to verify data integrity, such as confirming that an emergency vehicle request has not been tampered with. If a security vulnerability is discovered (such as key leakage), the data flow can be traced through the blockchain to quickly locate and isolate the affected nodes.

[0029] In one embodiment, the federated learning framework shares AI models across intersections and combines them with local training data to improve global prediction accuracy, including the following steps: Each temporary traffic light (intersection node) is loaded with a pre-trained basic AI model during deployment. This model has been initialized using public traffic datasets (such as urban road traffic flow and pedestrian behavior patterns) and has basic traffic prediction capabilities. The intersection node uses multimodal data (vehicle density, pedestrian numbers, emergency vehicle request records, etc.) uploaded from the perception layer, combined with local environmental parameters, such as school dismissal times and peak hours in commercial areas, to generate a labeled dataset. Input features: historical traffic density over the past 5 minutes, current weather conditions, and traffic light status at nearby intersections; Output objectives: Traffic flow forecast for each lane and pedestrian crossing demand probability for the next 10 minutes; The intersection nodes run local training tasks on edge computing devices and fine-tune the base model using labeled datasets, such as adjusting the attention mechanism weights of the gated Transformer to adapt the model to the unique traffic patterns of the intersection, such as the high proportion of pedestrians at intersections near schools during morning and evening rush hours and the high volume of trucks at intersections in industrial areas. During training, only the model parameters (weight matrix, bias terms) are retained, and the original data is always stored locally and not transmitted to external systems. After local training is completed, the intersection nodes encrypt and upload the updated model parameters (rather than the original data) to the central federated learning server (deployed in the traffic management center). Differential privacy techniques (such as adding Gaussian noise) are used during upload to further protect local data privacy (avoiding the inference of traffic flow values ​​from parameters). After receiving the model parameters from all intersection nodes, the federated server runs an aggregation algorithm. Weight calculation: Weights are assigned based on the amount of data at intersection nodes (such as the number of traffic density samples), with nodes having a larger amount of data contributing more; Anomaly detection: By statistically calculating the Euclidean distance of the weight matrix of each node through parameter distribution, abnormal nodes are identified and eliminated, such as those whose parameters deviate from the normal range due to sensor failure. The aggregated global model parameters generate a new "global baseline model," which integrates traffic pattern characteristics of multiple intersections, such as traffic flow patterns under different functional zones and weather conditions. The federated server distributes the aggregated global model parameters to all intersection nodes, overwriting the original local models. After receiving the global model, the intersection nodes combine it with the latest data they have stored locally and perform secondary fine-tuning, such as adjusting the output layer of the global model to match the traffic light phase configuration of the intersection, ensuring that the model has both global prediction capabilities and can adapt to local details. The updated model is used at intersection nodes to predict traffic flow and pedestrian demand in real time for the next 10 minutes (e.g., "Traffic flow from east to west on the main road will increase by 30% in the next 5 minutes"). The prediction results are then input into the dynamic timing engine to adjust the traffic light phases and extend the green light time for east to west. The prediction results are also sent back to the federated server as a reference for the next round of global model aggregation (intersections marked with "higher prediction errors during school dismissal time" will have their weight increased in subsequent aggregations). The federal server sets fixed periods or trigger conditions (detection of major traffic events or sudden weather changes) to initiate a new round of model aggregation and distribution, ensuring that the global model always reflects the latest traffic patterns (changes in traffic flow during holidays and the impact of newly opened roads). The raw data is always stored locally. The federated server only has access to the encrypted model parameters and cannot infer the traffic flow details at specific intersections. For example, it cannot determine from the parameters that "there are 100 cars at a certain intersection at 7 a.m.". The variance of the parameters of each node is calculated by monitoring the parameter distribution. If the parameters of a node are detected to deviate continuously from the global mean (due to sensor failure or malicious attack), its model upload permission is immediately suspended and it is marked as requiring manual inspection.

[0030] In one embodiment, the dynamic timing engine predicts short-term traffic trends based on the ARIMA algorithm and analyzes the current traffic flow in real time through a gated Transformer, including the following steps: Input historical traffic flow data (traffic density every 5 minutes in the past hour), eliminate non-stationarity (trend, seasonality) through difference operation, determine the difference order d; calculate the autocorrelation function and partial autocorrelation function, determine the autoregression order p and the moving average order q, and finally select the ARIMA model as ARIMA(p,d,q); Use maximum likelihood estimation to fit the model parameters (autoregressive coefficients). , , ..., Moving average coefficient , , ..., (Constant term c); Based on the fitted model, predict the traffic flow trend in the next h steps and output the predicted value. ; Calculate prediction error ( (Based on actual traffic flow), if the error exceeds the threshold (e.g., mean square error MSE > 10%), the model order is redefined and the parameters are adjusted. Core formula:

[0031] in, Let be the predicted traffic density at time t; The actual traffic density at time ti; The prediction error at time tj; , represents the autoregressive and moving average coefficients; c is a constant term (such as average traffic density). Input real-time traffic density data ( ), number of pedestrians ( ), environmental parameters (such as visibility) Weather conditions Data is mapped to vectors through a linear embedding layer. , dimension ; Calculate the query, key, and value matrix: , ,

[0032] in, , , The learnable parameter matrix; Calculate the attention weight A (scaled dot product attention):

[0033] in, The key vector dimension is used for scaling to prevent gradient vanishing; the attention output and the original input are fused through a gating mechanism.

[0034] Where σ is the Sigmoid function, represents the gating parameter, and ⊙ represents element-wise multiplication; The gated Transformer outputs a comprehensive feature vector of the current traffic flow. The data is input to the fully connected layer to generate real-time traffic flow analysis results (e.g., "The current traffic density on the main road from east to west is 20 vehicles / minute, and the probability of pedestrian crossing demand is 30%").

[0035] Where Q, K, and V are the query, key, and value matrices, respectively; A is the key vector dimension (used for scaling); A is the attention weight matrix (representing the importance of each time step / feature); V is the value matrix (containing original data information); ARIMA provides traffic flow trends for the next 10 minutes, while the gating Transformer analyzes the current traffic flow status. The results from both are fed into the dynamic timing engine. Based on the fluctuations in real-time traffic flow, such as a sudden event causing a surge in traffic flow, the output weight of the gating Transformer will be dynamically increased to ensure that traffic light adjustments both reference long-term trends and respond to immediate changes.

[0036] In one embodiment, the blockchain verification module receives an encrypted request from an emergency vehicle, verifies its identity via a smart contract, and then triggers a forced switch of the traffic lights to the priority phase, including the following steps: Emergency vehicles (ambulances, fire trucks) generate encrypted requests via onboard terminals or dedicated equipment, containing the following information: Identification information: Unique vehicle identifier (such as license plate number, rescue unit registration number); Location and Route: Current GPS coordinates, target location (e.g., hospital, fire scene), planned route (latitude and longitude sequence); Priority: The priority level corresponding to the type of rescue (e.g., "Level 1 Fire" or "Level 2 Medical Emergency"); Timestamp: The specific time the request was generated; The vehicle terminal sends the encryption request to the nearest temporary traffic light transmission layer node via the 5G / C-V2X communication module, enabling low-latency transmission; After receiving the request, the transport layer node encapsulates it into a blockchain transaction, which includes: Raw data: Plaintext information of emergency vehicle requests (identity, location, priority, etc.); Digital signature: A signature is generated using an asymmetric encryption algorithm to ensure that the transaction cannot be tampered with; Hash value: Calculate the SHA-256 hash of the original data, which serves as a unique identifier for the transaction; The encapsulated transaction is broadcast to all participating nodes (including other temporary traffic lights and traffic management center servers) via the blockchain network, forming a distributed ledger record. Each node independently stores a copy of the transaction, ensuring data redundancy and resistance to attacks. Blockchain nodes run pre-defined smart contracts (written in Solidity) to automatically perform verification: Identity legitimacy: By comparing with the whitelist pre-stored by the traffic management center (such as the license plate of an ambulance registered with the emergency center), it is confirmed whether the vehicle is a legitimate rescue unit; Request timeliness: Check if the timestamp is within the validity period (e.g., the difference between the request generation time and the current time is less than 5 minutes) to prevent expired or forged requests; Priority matching: Set priority thresholds based on the type of rescue (e.g., "Level 1 Fire" > "Level 2 Medical Emergency") to ensure that high-priority requests are processed first; Upon successful verification, the smart contract generates a "priority pass credential" (containing a hash value, verification result, and validity period) and triggers a traffic light adjustment command. If verification fails, such as when the vehicle is not on the whitelist or the request has expired, the smart contract directly rejects the request and records an alarm log. The transport layer node sends the "priority pass credential" generated by the smart contract to the platform layer dynamic timing engine via the 5G / C-V2X module, and the engine executes it immediately: Forced phase switching: Based on the planned route of emergency vehicles (such as "traveling east to west along the main road"), switch the traffic light in the target direction to green and extend the green light time; Conflict direction suppression: Temporarily switch the light to red for directions that conflict with emergency vehicle routes to avoid intersection congestion; The platform layer sends broadcast messages to vehicles near the emergency vehicle's path via the 5G / C-V2X module: traffic light status (e.g., "Eastbound green light has been extended to 60 seconds"), recommended speed, and abnormal warnings (e.g., "Ambulance is about to pass, please do not run the red light"). After the traffic light adjustment is completed, the transmission layer nodes package the execution results (such as "green light extended to 60 seconds" and "no conflicting vehicles") into blockchain transactions and send them back to the blockchain network, forming a closed-loop verification. All nodes update their ledger records to ensure data consistency. When requesting packaging, random noise is added to non-critical data (such as the specific vehicle model) to prevent the inference of detailed vehicle information from the blockchain data; if a node continues to send invalid requests (such as forging whitelisted vehicles), the blockchain network marks it as abnormal through the consensus mechanism and suspends its transaction processing authority until manual intervention is required for verification.

[0037] In one embodiment, the integrated user feedback entry point, after verification, dynamically optimizes timing, including the following steps: Users can access the "Traffic Feedback" module through the mobile app, select the feedback type (e.g., "Congestion," "Accident," "Traffic Light Malfunction"), and fill in the required information. Location: Automatically locate the current position or manually select the intersection; Time: Defaults to the current time, but can be manually adjusted (e.g., to report past traffic congestion). Description: Text input box (e.g., "Vehicles traveling from east to west are queuing up to 50 meters outside the intersection"). Evidence: You can choose to upload images (such as camera screenshots), videos (such as dashcam footage), or voice descriptions. The app performs initial screening of feedback, filtering out invalid content (such as blank descriptions or lack of location information) and automatically marking feedback as urgent, such as when the user selects the "Accident" type and marks it as "Requires Urgent Handling"; Feedback data is sent to the platform layer via the 5G / C-V2X module or Bluetooth Low Energy in the transport layer, and is encrypted with AES-256 to ensure that the data is not stolen or tampered with during transmission; emergency feedback is marked as high priority and transmitted through dedicated bandwidth; ordinary feedback (such as congestion) is marked as medium priority. After receiving feedback, the platform layer verifies it by combining data from the perception layer, historical records, and other user feedback: Comparison with sensor data: If a user reports "congestion in the eastbound direction", the system checks the camera and radar data at the intersection to confirm whether the traffic density exceeds the threshold (e.g., more than 20 vehicles / minute). Cross-validation with other user feedback: If multiple users report the same problem in the same location and at similar times, the credibility is increased; Historical behavior analysis: If a user has submitted false feedback multiple times in the past (such as false reporting of congestion), the system reduces the weight of their feedback; If the feedback involves an emergency, the platform layer can call a blockchain smart contract to automatically verify the user's identity (such as whether they are a registered driver) and the reasonableness of the feedback content (such as whether the accident description conforms to common patterns); The verified feedback is input into the dynamic traffic signal timing engine, and the engine executes according to the feedback type: Congestion feedback: Extend the green light time in the congested direction, or adjust the signal phases at adjacent intersections to divert vehicles; Accident feedback: Temporarily switch the traffic signals at the accident intersection to the "accident mode", such as extending the red light time in all directions to prioritize the passage of rescue vehicles; Traffic signal anomaly feedback: If a user reports that "the green light in a certain direction is not on", the system remotely diagnoses the hardware status of the traffic signal. If a fault is confirmed, it switches to the standby traffic signal mode, such as enabling temporary indicator lights; Frequent congestion feedback data (such as fixed congestion at a certain intersection from 17:00 to 18:00 every day) is incorporated into the federated learning framework as local training data to update the AI model at that intersection, improving the accuracy of subsequent traffic flow predictions, such as the model learning to "increase the green light time from east to west after 17:00"; The platform layer pushes a processing notice to the user through the mobile APP: Feedback status: "Verified and processed" or "Failed verification (evidence needs to be supplemented)"; Details of traffic signal timing adjustment: Such as "The green light from east to west has been extended to 45 seconds, expected to relieve congestion"; Reward mechanism: Users can obtain points for each valid feedback, and the points can be used to exchange for benefits such as parking coupons, incentivizing continuous participation. [[ID=2,2]]

[0038] The adjusted traffic signal timing parameters are sent down to the edge computing nodes through the transport layer. The nodes update the local traffic signal control logic and synchronize it to the cameras and radars in the sensing layer to ensure the consistency of sensor data and traffic signal timing status. For example, the camera adjusts the pedestrian detection area according to the new green light time; If a user frequently submits false feedback (such as false reporting of congestion), the system marks them as an "abnormal user" through behavior analysis (such as the feedback location not matching the actual GPS) and restricts their feedback permission; Personal information involved in the user's feedback, such as the license plate number, is desensitized, such as replaced with "Yue B****", only retaining the necessary location and event description.

[0039] A control method for an intelligent traffic light control system, including: Step 1: Real-time traffic data is collected by a multimodal sensor array (cameras, millimeter-wave radar, infrared sensors, and environmental sensors) deployed at the intersection, covering traffic density, pedestrian behavior, vehicle speed and distance, weather parameters, etc.; edge computing nodes clean the raw data (remove noise), extract features (identify the number of pedestrians and the direction of traffic flow), and adapt to the environment (such as enhancing the sensitivity of infrared sensors in low light conditions at night), generating structured data that is uploaded to the transmission layer, while key data (such as the location of emergency vehicles) is retained locally for rapid decision-making; Step 2: The 5G / C-V2X module transmits the structured data from the perception layer to the platform layer at high speed, supporting vehicle-road cooperation; Bluetooth Low Energy enables short-range interaction between the mobile app and traffic lights, such as users adjusting signal timing; the blockchain network encrypts and transmits emergency vehicle requests (such as ambulances) and verifies smart contracts (verifying whitelists and timeliness), and the data is managed through priority queues (high-priority data occupies dedicated bandwidth) to reduce latency, and AES-256 encryption is used to ensure end-to-end security; Step 3: Integrate multi-source data and share AI models across intersections using a federated learning framework (combining local training data to improve global prediction accuracy); the dynamic timing engine predicts short-term traffic trends based on the ARIMA algorithm, analyzes the current traffic flow in real time through a gating Transformer, and then optimizes phase switching (extending green light time) through reinforcement learning to minimize total delay time; the blockchain verification module receives encrypted requests, verifies identity through a smart contract, and then triggers a forced switch of the traffic lights to the priority phase, for example, extending the green light for ambulances to 60 seconds; Step 4: Control commands generated at the platform layer, such as phase switching and brightness adjustment, are sent down to the application layer for execution via the transmission layer. The traffic light hardware directly adjusts the phase and brightness. The mobile app displays the traffic light status, estimated waiting time, and emergency vehicle warnings in real time. AR navigation assistance displays a countdown and recommended speed through in-vehicle devices, ensuring precise execution of control commands and improving user interaction efficiency. Step 5: Integrate user feedback entry points (reporting congestion, accidents) through the mobile APP, and dynamically optimize timing (adjusting the weights of the federated learning model) after cross-validating the feedback content (comparing it with sensor data); frequently reported data (such as daily congestion at a certain intersection at 17:00) are incorporated into the federated learning framework to update the local AI model and improve the accuracy of subsequent predictions; the management backend monitors the status of traffic lights, supports historical data backtracking and remote firmware upgrades, forming a closed loop of "execution-feedback-optimization".

[0040] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations 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. An intelligent traffic light control system, characterized in that, Comprise: Sensing layer: Real-time collection of traffic data through multi-modal sensor array, camera recognition of vehicle flow, pedestrian and non-motor vehicle behavior, millimeter wave radar detection of vehicle speed and distance, infrared sensor coverage of low-light night scene, environmental sensor monitoring of weather parameters; edge computing node pre-processes raw data and dynamically adjusts signal light brightness and mode based on environmental data; processed structured data is uploaded to platform layer through transmission layer, while retaining key data locally for rapid decision-making; Transmission layer: Multi-channel communication network is constructed, 5G / C-V2X module transmits sensing layer data to platform layer at high speed, supporting vehicle-road cooperation; low-power Bluetooth performs short-distance interaction between mobile phone APP and signal light; blockchain network encrypts and transmits emergency vehicle request and verifies it; data is managed through priority queue to reduce delay, and AES-256 encryption is used to ensure data security; Platform layer: Integrates multi-source data and executes intelligent decision-making, federal learning framework shares AI model across intersections, improves global prediction accuracy combined with local training data; dynamic timing engine predicts short-term traffic trends based on ARIMA algorithm, analyzes current traffic flow in real time through gated Transformer, and optimizes phase switching through reinforcement learning to minimize total delay time; blockchain verification module receives encrypted emergency vehicle request, verifies identity through smart contract, triggers signal light to switch to priority phase, and extends green light time; finally, the generated control instructions are sent to the application layer through the transmission layer for execution; Application layer: Mobile phone APP receives platform layer instructions, supports user remote movement of signal light, adjustment of signal time or setting of temporary control mode; real-time display of signal light state, estimated waiting time and emergency vehicle warning, and integration of user feedback entry for dynamic optimization of timing after verification; AR navigation assistance shows signal light countdown, recommended speed and abnormal event prompt to the driver through the vehicle-mounted device; Management background monitors all signal light states, supports historical data backtracking and remote firmware upgrade.

2. The intelligent traffic light control system of claim 1, wherein, The edge computing node pre-processes the raw data and dynamically adjusts the signal light brightness and mode based on the environmental data, comprising the following steps: The edge computing node first receives raw data from the multi-modal sensor array, including real-time video stream of the camera; structured data of the millimeter wave radar including vehicle speed, distance, and lane occupancy rate; thermal imaging data of the infrared sensor in low-light night scene; environmental parameters of the environmental sensor including temperature, humidity, visibility, and illumination intensity; the data is synchronized through timestamp, aligning multi-source information in time dimension to provide a basis for subsequent fusion analysis; The edge computing node performs lightweight processing on raw data to reduce redundant data transmission to the platform layer; removes noise in camera video streams, corrects abnormal distance measurement values of millimeter wave radar due to multipath effects; filters false detection signals of infrared sensors due to environmental heat radiation; identifies vehicle types, pedestrian numbers and behaviors through lightweight target detection algorithms; calculates lane-level traffic density and average speed; marks the thermal imaging contours of pedestrians and animals, and judges the moving direction; quantifies visibility and light intensity, and classifies weather conditions; Based on the pre-processed environmental parameters, the edge computing node dynamically evaluates the signal light visibility requirement; When the infrared sensor detects a pedestrian or an animal in a low-light environment, the signal light is triggered to switch to "night mode"; when the light intensity sensor detects that the light intensity is less than 100 lux, the pedestrian green light time is automatically extended to ensure the safety of pedestrians crossing the street; when the visibility sensor detects that the visibility is less than 50 meters, the signal light switches to "fog mode"; when the millimeter wave radar detects that the lane-level traffic density decreases, the signal cycle is temporarily shortened; The edge computing node directly triggers the signal light adjustment instruction according to the environmental analysis result; dynamically adjusts the current of LED lamp beads through PWM technology for stepless brightness adjustment; calls the preset signal light display template; in extreme scenarios, the edge node temporarily extends the current phase green light time and synchronously reports the abnormal event to the platform layer, requesting subsequent global timing optimization; The edge computing node uploads the pre-processed structured data and local adjustment records to the platform layer through the transmission layer; optimizes the mapping relationship between environmental parameters and signal light adjustment strategies in combination with other intersection data; the platform layer verifies the effectiveness of the global algorithm in extreme scenarios according to the local adjustment records of the edge node, and dynamically corrects the timing parameters.

3. The intelligent traffic light control system of claim 1, wherein, The blockchain network encrypts and verifies the emergency vehicle request; data is managed through a priority queue to reduce delay, and AES-256 encryption is used to ensure data security, including the following steps: The emergency vehicle generates an encrypted request containing vehicle identity information, current location, target route and priority through the vehicle terminal or special equipment, and the request content is sent to the nearest temporary traffic signal light transmission layer node through the vehicle 5G / C-V2X module; after receiving the request, the transmission layer node packages the raw data into a blockchain transaction, generates a digital signature using an asymmetric encryption algorithm, making the request tamper-proof; the encrypted transaction is broadcast to all participating nodes through the blockchain network, forming a distributed ledger record; The blockchain node runs a smart contract to automatically verify the legality of the vehicle by comparing it with a preset whitelist; checks whether the request generation time is within the valid period, and sets a priority threshold according to the type of rescue to give priority to high-priority requests; After verification, the smart contract generates a "priority pass certificate" and triggers the signal light adjustment instruction; The transmission layer node sends the "priority pass certificate" verified to the platform layer through the 5G / C-V2X module, and the platform layer dynamic timing engine immediately adjusts the signal lamp phase and synchronously broadcasts the signal lamp change to nearby vehicles; after the signal lamp executes the adjustment, the execution result is returned to the blockchain network through the transmission layer, forming a closed-loop verification; The transmission layer classifies all uplink and downlink data and marks the priority: High priority: emergency vehicle request, real-time video stream, signal lamp adjustment instruction; Medium priority: structured sensor data, user feedback; Low priority: historical data return, system log; The transmission layer uses a "strict priority queue" algorithm, high-priority data occupies dedicated bandwidth, medium-priority data allocates resources in the remaining bandwidth according to the "weighted fair queue", and low-priority data is only transmitted when the bandwidth is idle, and dynamic discarding is supported; The transmission layer monitors network load in real time, and when congestion is detected, high-priority data remains unchanged, medium-priority data reduces transmission rate, and low-priority data suspends transmission until the network recovers; All uplink and downlink data are encrypted by AES-256 before transmission, and the perception layer data is encrypted on the edge computing node; platform layer instructions are encrypted in the management background; user feedback data is encrypted on the mobile phone APP side; The key management adopts a "hierarchical key system", the master key is generated by the traffic management center and replaced regularly; the session key is dynamically generated by the transmission layer node and used for single communication, which is automatically destroyed after expiration; the encrypted data is transmitted to the receiving end through the transmission layer network, and the receiving end uses the corresponding session key to decrypt the data and restore the original content; if decryption fails, the receiving end discards the data immediately and triggers an alarm; All encrypted data is recorded by the blockchain network to form an unalterable audit log, and the management background can query historical data transmission records at any time to verify data integrity; if a security vulnerability is found, the data flow is traced through the blockchain to quickly locate the affected nodes and isolate them.

4. The intelligent traffic light control system of claim 1, wherein, The federated learning framework shares AI models across intersections to improve global prediction accuracy with local training data, including the following steps: Each temporary traffic signal loads a pre-trained basic AI model when deployed, which has been initialized through public traffic data sets and has basic traffic prediction capabilities; the intersection node uses the multi-modal data uploaded by the perception layer to generate a labeled data set combined with local environmental parameters; the input features are historical 5-minute traffic density, current weather conditions, and nearby intersection signal lamp status; the output target is the traffic prediction value of each lane and the probability of pedestrian crossing demand within the next 10 minutes; The intersection node runs a local training task on the edge computing device, fine-tunes the basic model using the labeled data set, and makes the model adapt to the unique traffic patterns of the intersection; during training, only the model parameters are retained, and the original data is always stored locally and not transmitted externally; After the intersection node completes local training, the updated model parameters are encrypted and uploaded to the central federated learning server; differential privacy technology is used during upload to further protect local data privacy; after the federated server receives the model parameters of all intersection nodes, it runs an aggregation algorithm, assigns weights according to the data volume of the intersection nodes, and the nodes with larger data volume contribute more; through parameter distribution statistics, abnormal nodes are identified and excluded; The aggregated global model parameters generate a new "global benchmark model" that integrates the traffic pattern characteristics of multiple intersections; the federated server distributes the aggregated global model parameters to all intersection nodes, overwriting the original local models; after receiving the global model, the intersection nodes combine it with the latest data retained locally for secondary fine-tuning; The intersection nodes use the updated model to predict future 10-minute traffic and pedestrian demand in real time, and input the prediction results into the dynamic timing engine to adjust the signal light phases; the prediction results are also fed back to the federated server as a reference for the next round of global model aggregation; The federated server sets a fixed period or trigger condition to start a new round of model aggregation and distribution, ensuring that the global model always reflects the latest traffic patterns; The original data is always stored locally, and the federated server only accesses encrypted model parameters, making it impossible to reverse-engineer the specific intersection traffic details; through parameter distribution monitoring, if a node's parameters consistently deviate from the global mean, its model upload permission is suspended, and it is marked for manual inspection.

5. The intelligent traffic light control system of claim 1, wherein, The dynamic timing engine predicts short-term traffic trends based on the ARIMA algorithm and analyzes current traffic in real time using the Gated Transformer, including the following steps: Input historical traffic data and eliminate non-stationarity through difference operation to determine the difference order d; calculate the autocorrelation function and partial autocorrelation function to determine the autoregressive order p and the moving average order q, and finally select the ARIMA model as ARIMA(p, d, q); fitting model parameters using maximum likelihood estimation; based on the fitted model, predicting traffic trend in future h steps, outputting prediction value ; calculating prediction error for actual traffic, if error exceeds threshold, re-ordering model and adjusting parameters Core formula: wherein, is the predicted traffic density at time t; is the actual traffic density at time t-i; is the prediction error at time t-j; , is the autoregressive and moving average coefficient; c is the constant term; input real-time data traffic density , number of pedestrians , visibility , weather state ; mapping data into vectors by linear embedding layer , dimension ; Calculate the query Q, key K, and value V matrices: , , wherein, , , is a learnable parameter matrix; Calculate the attention weight A: where, is the key vector dimension, used to scale to prevent gradient vanishing; fuses the attention output with the original input through a gating mechanism: wherein σ is a Sigmoid function, is a gating parameter, and is an element-wise multiplication. The gated Transformer outputs a comprehensive feature vector of the current traffic flow which is input to a fully connected layer to generate a real-time traffic flow analysis result: wherein Q, K, V are query, key, value matrices respectively; is the key vector dimension; A is the attention weight matrix; V is the value matrix; ARIMA provides a 10-minute traffic trend, and Gated Transformer analyzes the current traffic state; the results of both are input into the dynamic timing engine; based on the volatility of real-time traffic, the output weight of Gated Transformer is dynamically increased, allowing signal light adjustments to reference long-term trends and respond to immediate changes.

6. The intelligent traffic light control system of claim 1, wherein, The blockchain verification module receives encrypted requests from emergency vehicles, verifies their identities through smart contracts, and triggers the signal light to switch to the priority phase, including the following steps: Emergency vehicles generate encrypted requests containing identity information, location and route, priority, and timestamp through vehicle terminals or dedicated devices; The vehicle terminal sends the encrypted request to the nearest temporary traffic signal light transmission layer node through the 5G / C-V2X communication module; after receiving the request, the transmission layer node encapsulates it as a blockchain transaction, including the original data, digital signature, and hash value; the encapsulated transaction is broadcast to all participating nodes through the blockchain network, forming a distributed ledger; each node independently stores a copy of the transaction, improving data redundancy and attack resistance. The blockchain node runs a preset smart contract, automatically compares with the pre-stored whitelist of the traffic management center, confirms whether the vehicle is a legal rescue unit, checks whether the timestamp is within the valid period to prevent expired or fake requests, sets a priority threshold according to the rescue type, and makes high-priority requests be processed first, and after verification, the smart contract generates a "priority pass certificate" and triggers a signal light adjustment instruction. If the verification fails, the smart contract directly rejects the request and records an alarm log. The transmission layer node sends the "priority pass certificate" generated by the smart contract to the platform layer dynamic timing engine through the 5G / C-V2X module, and the engine executes immediately according to the emergency vehicle planning path, switches the target direction signal light to green and extends the green light time, and temporarily switches the direction conflicting with the emergency vehicle path to red to avoid intersection congestion. The platform layer sends a broadcast message to vehicles near the emergency vehicle path through the 5G / C-V2X module: signal light state, recommended speed, and abnormal prompt. After the signal light adjustment is completed, the transmission layer node packages the execution result as a blockchain transaction and returns it to the blockchain network, and all nodes update the ledger record. When packaging the request, random noise is added to non-critical data. If a node continuously sends invalid requests, the blockchain network will mark it as abnormal through the consensus mechanism and suspend its transaction processing authority until manual intervention is checked.

7. The intelligent traffic light control system of claim 1, wherein, The integrated user feedback entry is dynamically optimized for timing after verification, The method comprises the following steps: The user enters the "traffic feedback" module through the mobile phone APP, selects the feedback type, and fills in the location, time, description, and evidence information. The APP performs preliminary screening on the feedback, filters out invalid content, and automatically marks the emergency feedback. The feedback data is sent to the platform layer through the 5G / C-V2X module and low-power Bluetooth of the transmission layer, and is encrypted by AES-256 to prevent data from being stolen or tampered with during transmission. The emergency feedback is marked as high priority and transmitted through a dedicated bandwidth, and the ordinary feedback is marked as medium priority. After receiving the feedback, the platform layer verifies it in combination with the sensing layer data, historical records, and other user feedback, compares it with sensor data, cross-verifies it with other user feedback, and analyzes historical behavior. If the feedback involves an emergency, the platform layer calls the blockchain smart contract to automatically verify the user's identity and the reasonableness of the feedback content. The verified feedback is input into the dynamic timing engine, and the engine executes according to the feedback type: Congestion feedback: extend the green light time of the congested direction or adjust the signal light phase of the adjacent intersection to divert vehicles; Accident feedback: temporarily switch the signal light of the accident intersection to "accident mode"; Signal light abnormal feedback: If the user reports that "the green light in a certain direction does not light up", the remote diagnosis of the signal light hardware state is performed, and if the fault is confirmed, the standby signal light mode is switched to. Frequent congestion feedback data is incorporated into the federal learning framework to update the AI model of the intersection as local training data; the platform layer pushes the processing notification feedback state to the user through the mobile APP, adjusts the details in time, and rewards the mechanism; the adjusted signal light timing parameters are issued to the edge computing node through the transmission layer, the node updates the local signal light control logic, and synchronizes to the camera and radar of the perception layer; If the user frequently submits false feedback, he will be marked as "abnormal user" through behavior analysis and his feedback rights will be restricted; The personal information involved in the user feedback is desensitized, and only the necessary location and event description are retained.

8. The intelligent traffic light control system control method according to claim 1, comprising the following steps: Step one: real-time collection of traffic data through multi-modal sensor arrays deployed at intersections, covering traffic density, pedestrian behavior, vehicle speed and distance, and weather parameters; edge computing nodes clean, extract features, and adapt to the environment, generate structured data uploaded to the transmission layer, while retaining key data locally for rapid decision-making; Step two: high-speed transmission of perception layer structured data to the platform layer through 5G / C-V2X modules, supporting vehicle-road cooperation; low-power Bluetooth for short-range interaction between mobile APP and signal lights; blockchain network for encrypted transmission and smart contract verification of emergency vehicle requests, data through priority queue management to reduce delay, and AES-256 encryption to ensure end-to-end security; Step three: integrate multi-source data, share AI models across intersections through federal learning framework; dynamic timing engine predicts short-term traffic trends based on ARIMA algorithm, analyzes current traffic in real time through gated Transformer, and optimizes phase switching through reinforcement learning to minimize total delay time; blockchain verification module receives encrypted requests, verifies identity through smart contract, and triggers signal light to switch to priority phase; Step four: control instructions generated by the platform layer are issued to the application layer through the transmission layer for execution, and the signal light hardware directly adjusts the phase and brightness; the mobile APP displays the signal light state, estimated waiting time, and emergency vehicle warning in real time; AR navigation assistance displays countdown and recommends speed through vehicle-mounted devices; Step five: integrate user feedback through the mobile APP, dynamically optimize timing after cross-verification of feedback content; frequent feedback data is incorporated into the federal learning framework to update the local AI model; the management background monitors the signal light state, supports historical data backtracking, and supports remote firmware upgrade.

Images