Traffic signal control dynamic decision-making system and method based on RAG enhancement

Abstract

The invention relates to a traffic signal control dynamic decision-making system and method based on RAG enhancement, and relates to the technical field related to intelligent traffic, and the system comprises a multi-source data collaborative collection and knowledge construction module which is used for obtaining real-time traffic perception data and constructing a multi-source traffic knowledge graph; the data preprocessing and knowledge retrieval module is used for generating a real-time scene vector and performing similarity retrieval; the RAG enhancement decision generation and optimization module is used for performing traffic signal control dynamic decision and simulation verification based on the candidate knowledge set; and the feedback optimization and knowledge updating module is used for constructing a scheme effect label and carrying out graph updating and model optimization. The technical problems that in the prior art, traffic signal control depends on fixed timing, multi-source historical knowledge cannot be effectively fused to deal with complex dynamic scenes, and consequently decision-making flexibility and adaptability are poor, and interpretability is weak are solved, and the technical effects of improving decision-making flexibility and adaptability, intersection passing efficiency and management level are achieved.

Description

A RAG-enhanced dynamic decision-making system and method for traffic signal control Technical Field

[0001] This invention relates to the field of intelligent transportation technology, specifically to a dynamic decision-making system and method for traffic signal control based on RAG enhancement. Background Technology

[0002] In urban traffic management, traffic signal control is a core means of regulating the flow of vehicles and pedestrians at intersections. The scientific nature of its decision-making directly affects road traffic efficiency and traffic safety. Traditional traffic signal control methods have significant limitations. On the one hand, fixed timing control relies on preset time periods and cannot respond to real-time traffic flow fluctuations, such as sudden traffic surges during morning and evening rush hours or sudden changes in traffic flow caused by temporary traffic control, which can easily lead to intersection congestion or waste of traffic resources. On the other hand, existing adaptive control is mostly based on single data such as traffic flow collected by loop detectors, lacking integrated consideration of historical traffic patterns, weather effects, special events such as sports events and exhibitions, resulting in one-sided decisions. For example, the green light time for pedestrians to cross the street is not extended when the road surface is slippery in rainy weather, or the traffic flow release order is not optimized in advance around commercial areas during holidays.

[0003] With the development of intelligent traffic sensing and big data technologies, high-definition cameras, microwave radars, and other equipment deployed at intersections can collect dynamic data such as traffic flow, vehicle speed, and queue length in real time. Meanwhile, traffic management departments accumulate massive amounts of static data, including historical traffic data, accident records, and meteorological correlation analysis. Currently, adaptive control often uses real-time data in isolation, failing to effectively activate the value of historical data. For example, while real-time data reflects the current traffic situation, it lacks long-term pattern support; historical data, although containing rich scenario experience, is difficult to directly adapt to dynamically changing real-time scenarios.

[0004] Therefore, current technologies suffer from technical problems such as traffic signal control relying on fixed timing, inability to effectively integrate multi-source historical knowledge to cope with complex dynamic scenarios, resulting in poor decision-making flexibility and adaptability, and weak interpretability. Summary of the Invention

[0005] This application provides a RAG-enhanced traffic signal control dynamic decision-making system and method, which solves the technical problems in the prior art where traffic signal control relies on fixed timing and cannot effectively integrate multi-source historical knowledge to cope with complex dynamic scenarios, resulting in poor decision-making flexibility and adaptability and weak interpretability. It achieves the technical effect of improving decision-making flexibility and adaptability, intersection traffic efficiency and management level.

[0006] This application provides a RAG-enhanced traffic signal control dynamic decision-making system. The system includes: a multi-source data collaborative acquisition and knowledge construction module for acquiring real-time traffic perception data and constructing a multi-source traffic knowledge graph; a data preprocessing and knowledge retrieval module for preprocessing the real-time traffic perception data to generate real-time scene vectors and performing similarity retrieval on the real-time scene vectors through the multi-source traffic knowledge graph to obtain a candidate knowledge set; a RAG-enhanced decision generation and optimization module for performing dynamic traffic signal control decisions and simulation verification based on the candidate knowledge set to generate a timing scheme; and a feedback optimization and knowledge update module for constructing scheme effect labels for the timing scheme and updating the knowledge graph and optimizing the model based on the scheme effect labels.

[0007] In a possible implementation, the multi-source data collaborative acquisition and knowledge construction module includes: a real-time traffic perception data acquisition module, used to obtain the real-time traffic perception data through a pre-deployed cluster of perception devices; and a multi-source traffic knowledge graph construction module, used to extract and store knowledge from historical traffic datasets to generate the multi-source traffic knowledge graph.

[0008] In a possible implementation, the data preprocessing and knowledge retrieval module includes: a real-time data preprocessing module, used to preprocess the real-time traffic perception data to obtain the real-time scene vector; a knowledge retrieval and adaptation module, used to perform similarity retrieval of the real-time scene vector in the multi-source traffic knowledge graph using a cosine similarity algorithm to obtain an associated knowledge set; and an engine filtering module, used to filter invalid knowledge from the associated knowledge set through a rule engine to obtain the candidate knowledge set.

[0009] In a possible implementation, the RAG enhanced decision generation and optimization module includes: a decision generation submodule, used to make dynamic decisions on traffic signal control based on the candidate knowledge set and determine a dynamic timing scheme; and a scheme verification and execution module, used to perform simulation verification and distribution based on the dynamic timing scheme and obtain the distributed timing scheme.

[0010] In a possible implementation, the decision generation submodule includes: a basic timing calculation module, used to calculate an initial timing scheme based on real-time lane saturation using the Webster method; a RAG enhancement and optimization module, used to input the initial timing scheme and the candidate knowledge set into a large language model, and obtain a knowledge fusion timing scheme through generative reasoning; and a final scheme output module, used to generate the dynamic timing scheme based on the knowledge fusion timing scheme.

[0011] In a possible implementation, the scheme verification and execution module includes: a simulation verification module, used to simulate the dynamic timing scheme through a micro traffic simulation model and obtain simulation verification results; and a signal transmission module, used to push the dynamic timing scheme to the intersection traffic lights through a PLC controller when the simulation verification results meet predetermined conditions, and to determine the transmitted timing scheme.

[0012] In a possible implementation, the feedback optimization and knowledge update module includes: an effect feedback collection module, used to collect traffic parameters and manual evaluation data after the implementation of the issued timing scheme, to form the scheme effect label; and a knowledge and model iteration module, used to update the map and optimize the model based on the scheme effect label.

[0013] This application also provides a RAG-enhanced dynamic decision-making method for traffic signal control. The method includes: acquiring real-time traffic perception data and constructing a multi-source traffic knowledge graph; preprocessing the real-time traffic perception data to generate real-time scene vectors, and performing similarity retrieval on the real-time scene vectors through the multi-source traffic knowledge graph to obtain a candidate knowledge set; performing dynamic decision-making and simulation verification for traffic signal control based on the candidate knowledge set to generate a timing scheme; constructing scheme effect labels for the timing scheme, and updating the knowledge graph and optimizing the model based on the scheme effect labels.

[0014] This application proposes a RAG-enhanced dynamic decision-making system and method for traffic signal control. The system comprises a multi-source data collaborative acquisition and knowledge construction module for acquiring real-time traffic perception data and constructing a multi-source traffic knowledge graph; a data preprocessing and knowledge retrieval module for generating real-time scene vectors and performing similarity retrieval; a RAG-enhanced decision generation and optimization module for dynamic decision-making and simulation verification of traffic signal control based on candidate knowledge sets; and a feedback optimization and knowledge update module for constructing scheme effect labels and updating the knowledge graph and optimizing the model. This addresses the technical problems in existing technologies, such as traffic signal control relying on fixed timing and the inability to effectively integrate multi-source historical knowledge to cope with complex dynamic scenarios, resulting in poor decision-making flexibility and adaptability, and weak interpretability. The system achieves the technical effect of improving decision-making flexibility and adaptability, intersection traffic efficiency, and management level. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings of the embodiments of this disclosure will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0016] Figure 1 is a schematic diagram of a traffic signal control dynamic decision-making system based on RAG enhancement provided in an embodiment of this application.

[0017] Figure 2 is a schematic flowchart of a dynamic decision-making method for traffic signal control based on RAG enhancement provided in an embodiment of this application.

[0018] Figure labeling: Multi-source data collaborative acquisition and knowledge construction module 10, data preprocessing and knowledge retrieval module 20, RAG enhanced decision generation and optimization module 30, feedback optimization and knowledge update module 40. Detailed Implementation

[0019] To further illustrate the technical means and effects adopted by the present invention in order to achieve the intended purpose, the following detailed description is provided in conjunction with the accompanying drawings and preferred embodiments, based on the specific implementation methods, structures, features and effects of the present invention.

[0020] This application provides a traffic signal control dynamic decision-making system based on RAG enhancement, as shown in Figure 1. The system includes:

[0021] The multi-source data collaborative acquisition and knowledge construction module 10 is used to acquire real-time traffic perception data and construct a multi-source traffic knowledge graph.

[0022] Furthermore, the specific configuration of the multi-source data collaborative acquisition and knowledge construction module 10 also includes a real-time traffic perception data acquisition module, used to obtain the real-time traffic perception data through a pre-deployed cluster of perception devices; and a multi-source traffic knowledge graph construction module, used to extract and store knowledge from historical traffic datasets to generate the multi-source traffic knowledge graph.

[0023] Preferably, the real-time traffic perception data acquisition module is used to collect multimodal traffic status data from the physical world in real time. Specifically, a cluster of perception devices is deployed at each entrance lane, exit lane, and pedestrian crossing at the intersection to simultaneously collect multi-dimensional dynamic traffic flow data. High-definition cameras with a frame rate of 25-30fps and a resolution of ≥1080P collect visual data such as vehicle queue length, pedestrian density, and non-motorized vehicle flow. Microwave radar with a detection distance of 0-150m and a speed measurement error of ≤2km / h collects motion data such as instantaneous vehicle speed and vehicle flow gap. Geomagnetic detectors with a response time of ≤0.1s collect single-lane traffic flow data. All perception devices synchronously generate timestamps with a time deviation of ≤50ms and associate device location information, such as latitude and longitude and lane number, and transmit the data to the data processing center in real time through edge computing nodes.

[0024] Preferably, the multi-source traffic knowledge graph construction module is used to structurally extract and store experience and knowledge in the traffic field from history and rules. Specifically, it extracts key information from historical traffic data, including historical timing schemes, including optimal phase durations under different time periods and weather conditions; the occurrence time, causes, and timing adjustments for congestion events; special scenario tags, such as school drop-off and pick-up times around schools and peak holiday periods in commercial areas; and supplements weather-related rules from public data, such as a 20% increase in pedestrian crossing time during heavy rain, as well as traffic regulations, such as the green light interval setting for yielding to pedestrians when turning right. A graph database is used to construct the multi-source traffic knowledge graph. Nodes include "scenario type," "traffic parameters," "timing schemes," and "influencing factors," while edges are defined as relationships such as "causal association," "adapted scenario," and "optimization direction," forming a structured traffic knowledge graph that expresses causal and adaptation relationships in the traffic field.

[0025] The data preprocessing and knowledge retrieval module 20 is used to preprocess the real-time traffic perception data, generate real-time scene vectors, and perform similarity retrieval on the real-time scene vectors through the multi-source traffic knowledge graph to obtain a candidate knowledge set.

[0026] Furthermore, the specific configuration of the data preprocessing and knowledge retrieval module 20 also includes: a real-time data preprocessing module, used to preprocess the real-time traffic perception data to obtain the real-time scene vector; a knowledge retrieval and adaptation module, used to perform similarity retrieval of the real-time scene vector in the multi-source traffic knowledge graph using a cosine similarity algorithm to obtain an associated knowledge set; and an engine filtering module, used to filter invalid knowledge from the associated knowledge set through a rule engine to obtain the candidate knowledge set.

[0027] Preferably, the real-time data preprocessing module is used to transform heterogeneous real-time sensing data collected by front-end sensors into standardized real-time scene vectors. Specifically, it preprocesses real-time traffic sensing data such as video frames, radar point clouds, and geomagnetic pulse signals, including applying digital filters such as Gaussian filtering and Kalman filtering to eliminate random noise and outliers in the data, using the YOLOv8 computer vision model to identify and count the number of vehicles, pedestrians, and non-motorized vehicles in the image, smoothing outliers from radar and geomagnetic data through Kalman filtering, and calculating key traffic parameters such as lane saturation, average speed, and queue length. Here, lane saturation is the ratio of actual traffic flow to traffic capacity. The preprocessed parameters such as traffic flow, speed, queue length, current weather, pedestrian density, and time period are aligned and normalized. Current weather and time period information are obtained through API, and a real-time scene feature vector with a dimension of 512 is constructed to represent the current traffic status, weather, and time at the intersection.

[0028] Preferably, the knowledge retrieval and adaptation module utilizes the generated real-time scene vectors to quickly identify the most relevant historical experiences from a vast knowledge graph. Specifically, the historical scene descriptions in the knowledge graph are pre-converted into vectors of the same dimension and stored in a vector database. The cosine similarity between the real-time scene vector and each historical scene vector in the knowledge base is calculated. The similarity in direction is determined by measuring the cosine of the angle between the two vectors in space. The closer the value is to 1, the more similar the scenes are. Based on a set similarity threshold, such as >0.85, all matching historical scenes and their associated timing schemes, congestion causes, optimization rules, and other knowledge are returned, forming a set of associated knowledge adapted to the current scene.

[0029] Preferably, the engine filtering module is used to filter invalid knowledge from the associated knowledge set through the rule engine, ensuring that the knowledge used for final decision-making is reasonable, safe, and effective. Specifically, multiple business logics and constraints are pre-set in the rule engine. For example, highway timing schemes are excluded from urban side road scenarios, optimization schemes for main road intersections are not applicable to side roads near school gates, rules involving closing a certain phase in historical schemes are prohibited from being triggered during weekday morning rush hours, and rules for ensuring emergency access are prioritized when current GPS data shows that an ambulance has passed. Then, each piece of knowledge in the associated knowledge set is compared with the predefined rules in the rule engine, retaining knowledge that meets all current constraints, and marking knowledge that violates any rule as invalid and removing it, ultimately obtaining a finely filtered candidate knowledge set that can be directly used to assist decision-making.

[0030] The RAG enhanced decision generation and optimization module 30 is used to perform dynamic decision-making and simulation verification of traffic signal control based on the candidate knowledge set, and generate a timing scheme for distribution.

[0031] Furthermore, the specific configuration of the RAG enhanced decision generation and optimization module 30 also includes a decision generation submodule, used to make dynamic decisions on traffic signal control based on the candidate knowledge set and determine the dynamic timing scheme; and a scheme verification and execution module, used to perform simulation verification and distribution according to the dynamic timing scheme and obtain the distributed timing scheme.

[0032] Furthermore, the specific configuration of the decision generation submodule also includes: a basic timing calculation module, used to calculate the initial timing scheme based on real-time lane saturation using the Webster method; a RAG enhancement and optimization module, used to input the initial timing scheme and the candidate knowledge set into a large language model, and obtain a knowledge fusion timing scheme through generative reasoning; and a final output scheme module, used to generate the dynamic timing scheme based on the knowledge fusion timing scheme.

[0033] Preferably, the RAG-enhanced decision generation and optimization module is the component that executes the core decision-making function. Based on the retrieved candidate knowledge set, it generates an executable signal timing scheme through artificial intelligence generation and digital simulation verification. The decision generation submodule is used to make dynamic decisions on traffic signal control based on the candidate knowledge set. Specifically, it adopts the classic traffic signal timing theory Webster's method, calculates based on key traffic parameters such as real-time lane saturation, and outputs an initial timing scheme, including the basic green light duration and phase sequence for each phase, ensuring that the scheme conforms to the basic principles of traffic engineering. The initial timing scheme is then input into a large language model to perform generative reasoning, that is, to retrieve specific historical experience knowledge, such as "pedestrian crossing time should be increased on rainy days" and "school district..." The system integrates and adjusts the initial timing scheme by adding a dedicated pedestrian phase during school dismissal time, thereby generating modification instructions or reasons for the initial scheme. For example, "extend the green light duration for north-south pedestrians by 3 seconds based on the optimization rules for rainy weather in the knowledge entry," thus determining the optimized timing scheme based on knowledge integration and the corresponding adjustment reasons. Finally, it integrates the adjustment instructions output by the large language model to modify the parameters of the initial timing scheme, forming a complete and accurate dynamic timing scheme that includes the phase order, the green light duration for each phase (accurate to 0.5 seconds), the yellow light duration (fixed at 3 seconds), and the red light interval. It also includes reasons for the scheme's adaptation, such as "due to the saturation of 0.82 at the east entrance, the green light is extended by 4 seconds; combined with the rainy weather rules, the pedestrian green light is increased by 2 seconds."

[0034] Furthermore, the specific configuration of the scheme verification and execution module also includes a simulation verification module, used to simulate the dynamic timing scheme through a micro traffic simulation model and obtain simulation verification results; and a signal transmission module, used to push the dynamic timing scheme to the intersection traffic lights through a PLC controller when the simulation verification results meet predetermined conditions, and to determine the transmitted timing scheme.

[0035] Preferably, the scheme verification and execution module is used to perform simulation verification and distribution based on the dynamic timing scheme. That is, before the scheme is distributed to the real traffic lights, its safety and effectiveness are tested in a virtual environment. Specifically, the dynamic timing scheme and the real-time traffic flow data of the current intersection are input into VISSIM or SUMO micro traffic simulation software for simulation. Rapid simulation and extrapolation are performed in the digital twin environment, such as simulating the traffic operation conditions of the next 10 minutes. The simulation engine predicts and outputs key traffic efficiency indicators, such as average intersection delay, queue length, and throughput capacity, to obtain simulation verification results. Next, the simulation verification results are compared with the predicted effects of executing the current scheme or the historical benchmark scheme. If the indicators are better than the current scheme, such as a delay reduction of ≥10%, then the scheme is executed; otherwise, it is returned to be re-optimized. The final dynamic timing scheme that has passed the verification is safely and smoothly issued to the programmable logic controller (PLC) at the intersection through the industrial control protocol. The PLC controller pushes the dynamic timing scheme to the traffic lights at the intersection, determines the issued timing scheme, and switches the traffic lights according to the phase sequence and duration of the new scheme to achieve smooth switching of signal timing, avoid sudden interruption of green light, and ensure that the switching transition time is ≤1 second to ensure traffic safety.

[0036] The feedback optimization and knowledge update module 40 is used to construct the scheme effect tags of the distribution timing scheme, and to perform graph updates and model optimization based on the scheme effect tags.

[0037] Furthermore, the specific configuration of the feedback optimization and knowledge update module also includes an effect feedback collection module, used to collect traffic parameters and manual evaluation data after the implementation of the issued timing scheme, and form the scheme effect label; and a knowledge and model iteration module, used to update the map and optimize the model based on the scheme effect label.

[0038] Preferably, the feedback optimization and knowledge update module uses multi-dimensional feedback to quantitatively and qualitatively evaluate the actual effect of the issued and implemented timing schemes. Specifically, the effect feedback collection module generates multi-dimensional effect scores for each implemented timing scheme. Specifically, it uses real-time traffic sensing equipment to collect traffic parameters within a preset evaluation period after the implementation of the timing scheme, including vehicle delay time, number of stops, queue length, traffic volume, and lane occupancy. Feedback from traffic management personnel is received and structuredly entered through the human-computer interaction interface of the traffic management center. Feedback typically uses standardized options, such as a slider to select "satisfied". The system allows users to select preset issues, including limited free text such as insufficient pedestrian crossing time or congested left-turn traffic. It integrates objective parameters with subjective evaluations, applying predefined assessment rules to generate final solution effectiveness labels. For example, if "average delay reduction ≥ 10%" and "no negative human feedback," the label is "Excellent," and the solution is recorded as effective. If "delay increase or decrease < 5%" or "one or more critical negative feedbacks are received," the label is "Poor," and the solution is recorded as ineffective. Other cases are rated "Average." The final output is a structured solution effectiveness label, linked to the corresponding solution, scenario, and original data.

[0039] Preferably, the knowledge and model iteration module is used to dynamically update and optimize the knowledge graph and model based on the solution effect labels. Specifically, for cases labeled "excellent", effective timing solutions, new scenario descriptions, and optimization rules are added to the knowledge graph. This includes structuring and permanently storing the successful "scenario features - decision knowledge - execution solution - good results" association in the multi-source traffic knowledge graph. For example, a new scenario node "summer rainstorm - evening rush hour - commercial district intersection" is added, and a "verified effective" relationship edge is established with the specific phase adjustment solution node adopted this time. For cases labeled "medium" or "poor", the reasons are analyzed and labeled. If a certain historical knowledge is not applicable in this scenario, restrictive annotations are added to the knowledge node or its weight is adjusted. If a completely new failure scenario is found, such as "temporary road construction", a new rule node is created for future rule filtering. Every two weeks or accumulating 100 new cases, labeled feedback data is used as a fine-tuning training set to update the parameters of the large language model and retrieval rules, improving the accuracy of knowledge matching and the rationality of solution generation. This ensures that the traffic efficiency of timing solutions in complex scenarios is improved by ≥15%. The final output is an updated and richer multi-source traffic knowledge graph, as well as a new version of the large language model and embedded model parameters with improved performance. This ensures improved traffic efficiency, reduced management costs, and high adaptability to the needs of multiple scenarios, providing support for the construction of a smart traffic signal control network.

[0040] In the preceding text, a traffic signal control dynamic decision-making system based on RAG enhancement according to an embodiment of the present invention was described in detail with reference to FIG1. ​​Next, a traffic signal control dynamic decision-making method based on RAG enhancement according to an embodiment of the present invention will be described with reference to FIG2. A traffic signal control dynamic decision-making method based on RAG enhancement, as shown in FIG2, includes: acquiring real-time traffic perception data and constructing a multi-source traffic knowledge graph; preprocessing the real-time traffic perception data to generate real-time scene vectors, and performing similarity retrieval on the real-time scene vectors through the multi-source traffic knowledge graph to obtain a candidate knowledge set; performing dynamic decision-making and simulation verification of traffic signal control based on the candidate knowledge set to generate a timing scheme; constructing scheme effect labels for the timing scheme, and updating the knowledge graph and optimizing the model based on the scheme effect labels.

[0041] In one possible implementation, the RAG-enhanced traffic signal control dynamic decision-making method further includes: a real-time traffic perception data acquisition module, used to obtain the real-time traffic perception data through a pre-deployed cluster of perception devices; and a multi-source traffic knowledge graph construction module, used to extract and store knowledge from historical traffic datasets to generate the multi-source traffic knowledge graph.

[0042] In one possible implementation, the RAG-enhanced traffic signal control dynamic decision-making method further includes: a real-time data preprocessing module for preprocessing the real-time traffic perception data to obtain the real-time scene vector; a knowledge retrieval and adaptation module for performing a cosine similarity algorithm on the real-time scene vector in the multi-source traffic knowledge graph to obtain an associated knowledge set; and an engine filtering module for filtering invalid knowledge from the associated knowledge set using a rule engine to obtain the candidate knowledge set.

[0043] In one possible implementation, the RAG-enhanced traffic signal control dynamic decision-making method further includes: a decision generation submodule, used to make dynamic traffic signal control decisions based on the candidate knowledge set and determine a dynamic timing scheme; and a scheme verification and execution module, used to perform simulation verification and distribution based on the dynamic timing scheme and obtain the distributed timing scheme.

[0044] In one possible implementation, the RAG-enhanced traffic signal control dynamic decision-making method further includes: a basic timing calculation module, used to calculate an initial timing scheme based on real-time lane saturation using the Webster method; a RAG enhancement optimization module, used to input the initial timing scheme and the candidate knowledge set into a large language model, and obtain a knowledge fusion timing scheme through generative reasoning; and a final scheme output module, used to generate the dynamic timing scheme based on the knowledge fusion timing scheme.

[0045] In one possible implementation, the RAG-enhanced traffic signal control dynamic decision-making method further includes: a simulation verification module, used to simulate the dynamic timing scheme through a micro-traffic simulation model and obtain simulation verification results; and a signal distribution module, used to push the dynamic timing scheme to the intersection traffic lights through a PLC controller when the simulation verification results meet predetermined conditions, and to determine the distributed timing scheme.

[0046] In one possible implementation, the RAG-enhanced traffic signal control dynamic decision-making method further includes: an effect feedback acquisition module, used to collect traffic parameters and manual evaluation data after the issuance of the timing scheme, and form the scheme effect label; and a knowledge and model iteration module, used to update the map and optimize the model based on the scheme effect label.

[0047] The RAG-enhanced traffic signal control dynamic decision-making system provided in this embodiment of the invention can execute the RAG-enhanced traffic signal control dynamic decision-making method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0048] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A dynamic decision-making system for traffic signal control based on RAG enhancement, characterized in that, The system includes: a multi-source data collaborative acquisition and knowledge construction module, used to acquire real-time traffic perception data and construct a multi-source traffic knowledge graph; a data preprocessing and knowledge retrieval module, used to preprocess the real-time traffic perception data, generate real-time scene vectors, and perform similarity retrieval on the real-time scene vectors through the multi-source traffic knowledge graph to obtain a candidate knowledge set; a RAG enhanced decision generation and optimization module, used to perform dynamic decision-making and simulation verification of traffic signal control based on the candidate knowledge set, and generate a timing scheme; and a feedback optimization and knowledge update module, used to construct scheme effect labels for the timing scheme, and perform graph updates and model optimization based on the scheme effect labels.

2. The RAG-enhanced traffic signal control dynamic decision-making system as described in claim 1, characterized in that, The multi-source data collaborative acquisition and knowledge construction module includes: a real-time traffic perception data acquisition module, used to obtain the real-time traffic perception data through a pre-deployed cluster of perception devices; and a multi-source traffic knowledge graph construction module, used to extract and store knowledge from historical traffic datasets to generate the multi-source traffic knowledge graph.

3. The RAG-enhanced traffic signal control dynamic decision-making system as described in claim 1, characterized in that, The data preprocessing and knowledge retrieval module includes: a real-time data preprocessing module, used to preprocess the real-time traffic perception data to obtain the real-time scene vector; a knowledge retrieval and adaptation module, used to perform similarity retrieval of the real-time scene vector in the multi-source traffic knowledge graph using a cosine similarity algorithm to obtain an associated knowledge set; and an engine filtering module, used to filter invalid knowledge from the associated knowledge set through a rule engine to obtain the candidate knowledge set.

4. The RAG-enhanced traffic signal control dynamic decision-making system as described in claim 1, characterized in that, The RAG enhanced decision generation and optimization module includes: a decision generation submodule, used to make dynamic decisions on traffic signal control based on the candidate knowledge set and determine a dynamic timing scheme; and a scheme verification and execution module, used to perform simulation verification and distribution based on the dynamic timing scheme and obtain the distributed timing scheme.

5. The RAG-enhanced traffic signal control dynamic decision-making system as described in claim 4, characterized in that, The decision generation submodule includes: a basic timing calculation module, used to calculate an initial timing scheme based on real-time lane saturation using the Webster method; a RAG enhancement and optimization module, used to input the initial timing scheme and the candidate knowledge set into a large language model and obtain a knowledge fusion timing scheme through generative reasoning; and a final scheme output module, used to generate the dynamic timing scheme based on the knowledge fusion timing scheme.

6. The RAG-enhanced traffic signal control dynamic decision-making system as described in claim 4, characterized in that, The scheme verification and execution module includes: a simulation verification module, used to simulate the dynamic timing scheme through a micro traffic simulation model and obtain simulation verification results; and a signal transmission module, used to push the dynamic timing scheme to the intersection traffic lights through a PLC controller when the simulation verification results meet predetermined conditions, and to determine the transmitted timing scheme.

7. The RAG-enhanced traffic signal control dynamic decision-making system as described in claim 1, characterized in that, The feedback optimization and knowledge update module includes: an effect feedback collection module, used to collect traffic parameters and manual evaluation data after the implementation of the issued timing scheme, and form the scheme effect label; and a knowledge and model iteration module, used to update the map and optimize the model based on the scheme effect label.

8. A dynamic decision-making method for traffic signal control based on RAG enhancement, characterized in that, The method is applied to a traffic signal control dynamic decision-making system based on RAG enhancement as described in any one of claims 1-7. The method includes: acquiring real-time traffic perception data and constructing a multi-source traffic knowledge graph; preprocessing the real-time traffic perception data to generate real-time scene vectors, and performing similarity retrieval on the real-time scene vectors through the multi-source traffic knowledge graph to obtain a candidate knowledge set; performing dynamic decision-making and simulation verification for traffic signal control based on the candidate knowledge set to generate a timing scheme; constructing scheme effect labels for the timing scheme, and updating the knowledge graph and optimizing the model based on the scheme effect labels.

Images