Intelligent traffic signal control method, system and program product based on Agent-ARIMA

Abstract

The invention discloses an intelligent traffic signal control method and system based on Agent-ARIMA, and a program product. The intelligent traffic signal control method comprises the steps of calculating the shortest green light time of each intersection direction, distributing the green light time of each direction, distributing the green light time of left-turn and straight lanes, adjusting the traffic cycle, calculating the time weight offset of multiple intersections, and predicting the traffic flow in the future by using an ARIMA model. And inputting the prediction time of the ARIMA algorithm into the LLM-Agent agent for time prediction, and the like. The method is suitable for different city scales, time distribution of the traffic lights is dynamically adjusted through accurate prediction and analysis of the traffic flow, the traffic jam condition of a large city can be optimized, and the traffic efficiency of a small city can also be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent transportation data processing, and more specifically, to an intelligent traffic signal control method, system and program product based on Agent-ARIMA. Background Art

[0002] Traffic congestion has become a key issue affecting human society and the environment. With the accelerating pace of urban migration, the rapid expansion of the urban population has further exacerbated this challenge. Against the backdrop of highly developed technology today, traffic management still faces many challenges, among which the time allocation of traffic lights is unreasonable in many areas. Optimizing traffic signal control (TSC) in this trend is a key research area in the field of intelligent transportation. Efficient traffic signal control can not only bring huge economic development and environmental protection benefits, but also improve the efficiency of social operation. However, effectively managing traffic signals in complex and changeable urban roads remains a daunting task.

[0003] Currently, the main traffic signal control systems applied in China, including the green signal ratio, cycle, and phase difference optimization technology SCOOT (Split Cycle Offset Optimizing Technique) and the Sydney Coordinated Adaptive Traffic System SCATS, rely on detectors installed at intersections and use classical traffic signal control theories and algorithms for regional coordinated timing calculation and optimization. However, in actual applications, due to the poor working environment of the detectors and frequent road construction, the detectors are in a poor state of repair. Signal control systems without detector-provided data can only be used in a degraded mode, running the pre-determined traffic light cycle timing scheme within a single signal machine. In addition, most existing systems are closed systems and cannot accept traffic flow detection information provided by other detection means. The signal optimization algorithms of existing systems do not well combine the latest research results of intelligent transportation, and the evaluation of the current traffic control system does not well utilize big data analysis means, etc.

[0004] In addition, there are currently traffic signal control methods based on reinforcement learning. Although reinforcement learning-based methods have shown excellent performance in various traffic scenarios. However, reinforcement learning-based methods also have some disadvantages. First, their generalization ability is limited, especially when migrating to larger road networks or in very uncommon situations (such as extremely high traffic conditions), because their training data only covers limited traffic situations. In addition, due to the black box nature of deep neural networks, reinforcement learning (RL)-based methods lack interpretability, making it difficult to explain why they take control actions under specific traffic conditions.

[0005] Therefore, developing an intelligent traffic signal control method and system suitable for different city sizes or specific traffic conditions is a technical problem that needs to be solved urgently. Summary of the invention

[0006] Due to the problems existing in the prior art, the present invention uses a large-scale language model as an intelligent agent and combines it with the ARIMA adaptive algorithm to propose a set of intelligent traffic signal control methods, systems and program products suitable for cities of different sizes. The system can dynamically adjust the time allocation of traffic lights by accurately predicting and analyzing the traffic flow, which can not only optimize the traffic congestion in large cities, but also improve the traffic efficiency in small cities, and provide more universal and refined solutions to rapidly changing traffic needs.

[0007] To achieve the above object, in a first aspect, the present invention provides an intelligent traffic signal control method based on Agent-ARIMA, comprising the following steps:

[0008] Step S1, calling the map API (Application Programming Interface) to obtain the congestion distance, average driving speed, and traffic volume of the left turn lane and the straight lane of each road section of the intersection in real time;

[0009] Step S2: Calculate the maximum time for all vehicles on the two intersections to pass the traffic lights , set the minimum green light duration , Enable pedestrians to cross the zebra crossing;

[0010] Step S3: Green light time allocation at intersection: green light cycle length According to the green light time required for each direction and OK, the formula is ;

[0011] Step S4, Time Allocation between Left-Turn Lane and Straight-Through Lane: When the required green light time for the x direction is , and the required green light time for the y direction is , ensure that and ; If this condition is not met for any direction, adjust the green light duration for the next cycle; If one side is less than , then set the green light time for this side to in the next cycle;

[0012] Step S5, Dynamic Adjustment of Traffic Cycle: Calculate the actual green light time for each direction in the next cycle based on the traffic flow ratio of each direction; The green light duration for the ab direction and the green light duration for the cd direction are respectively: ;

[0013] Step S6, Weight Offset Calculation for Multi-Intersection Time: Obtain the road section information, the required green light time for the ac road, and the required green light time for the bd road to obtain the distance matrix between each intersection; Calculate the new green light time allocation ratio based on various factors including the traffic congestion degree and traffic flow in two directions; Adjust the green light time ratio by calculating the weight range ;

[0014] Step S7, According to the ARIMA (Autoregressive Integrated Moving Average Model) algorithm, use the time series to fit the ARIMA model, and predict the future time period based on the model to obtain the predicted value ; The ARIMA model can be expressed as: ;

[0015] Among them, is the constant term, is the coefficient of the autoregressive term, is the coefficient of the moving average term, is the error term, usually regarded as white noise, p: the order of the autoregressive model, indicating the linear relationship between the current value and the past p values; q: the order of the moving average model, indicating the linear relationship between the current value and the past q prediction errors;

[0016] Step S8: Determine whether the traffic conditions at the intersection can be correctly predicted. An intersection with a mean absolute percentage error less than 10% is considered to be correctly predictable, and such intersections are called "simple intersections"; if the error is greater than 10%, such intersections are called "complex intersections"; if the intersection is a complex intersection, the signal control time series predicted for the intersection based on the ARIMA model is used as input data and input into the LLM-Agent signal control agent to predict the signal time; the LLM-Agent signal control agent is a large language model fine-tuned and trained based on traffic data.

[0017] Step S9: Obtain the latest road conditions after a fixed time interval, and update the time of the two types of intersections, namely "simple intersections" and "complex intersections", according to the latest road conditions.

[0018] The intelligent traffic signal control method first calculates and adjusts the time cycle of the traffic signal in real time, then combines real-time data and the ARIMA model to predict the future time period, and then uses the result predicted by the ARIMA model as the input of a large language model fine-tuned and trained based on traffic data to obtain a high-precision prediction of the real-time traffic flow and an optimized control of the signal time.

[0019] Further, in step S6, the road segment information includes but is not limited to the congestion distance, average vehicle speed, and congestion status.

[0020] Further, in step S6, the distance between intersections can be used to calculate the time for the traffic flow to reach the next intersection based on the current average vehicle speed of the road segment. Calculate the time for the upstream intersection to reach the current intersection. , where S is the distance between two adjacent upstream and downstream intersections, v is the vehicle flow speed, and it is saved as an array; calculate the current time , and determine whether it is less than the time required to arrive plus the time to obtain data , that is seconds. If it is greater, then keep this time; if it is insufficient, take the green light time of the previous time period on the same road as the weight to generate an array and store it; transfer to The time ratio is called the offset weight, and the weight calculation formula is: , and make a duration transfer adjustment for and .

[0021] Further, in step S8, the output of the ARIMA model is the predicted signal light durations arranged in chronological order, and each time point contains the corresponding signal light duration information; after being preprocessed and standardized, the predicted signal light durations arranged in chronological order are input into the time series analysis sub-module of the LLM-Agent signal light control agent in the form of embedding vectors or direct numerical values.

[0022] Further, in step S8, the steps for constructing the LLM model fine-tuned based on traffic data include: First, construct traffic state observation features and collect traffic state observation data at traffic intersections; Second, construct model prompts enhanced with common sense knowledge to form a prompt integrating common sense knowledge for guiding the LLM model to infer the optimal traffic signal configuration for the next time slice; Finally, the LLM model uses the constructed prompt for the analysis and reasoning decision-making process and then makes a decision.

[0023] Further, in step S9, for "simple intersections", repeat steps S1 to S7, and for "complex intersections", repeat steps S1 to S9.

[0024] In a second aspect, the present invention provides an intelligent traffic signal control system based on Agent-ARIMA for implementing the intelligent traffic signal control method based on Agent-ARIMA as described above, including:

[0025] A road section information collection module that collects the congestion distance, average driving speed, traffic flow of the left-turn lane and the straight-through lane at each road section of the crossroads.

[0026] An intersection signal light time allocation module that calculates the maximum time for all vehicles to pass through the traffic lights on two intersecting roads , sets the shortest green light duration , conducts intersection green light time allocation, left-turn lane and straight-through lane time allocation, and dynamic adjustment of the traffic cycle; based on a single intersection, performs weight offset for multi-intersection time allocation.

[0027] An ARIMA model that uses time series to fit the ARIMA model according to the ARIMA algorithm and predicts the future time period based on the ARIMA model to obtain predicted values ;

[0028] An LLM-Agent signal light control agent, a large language model fine-tuned based on traffic data; for complex intersections, the signal light control time series predicted based on the ARIMA model is used as the input data of the LLM-Agent signal light control agent to predict the signal light time.

[0029] In a last aspect, the present invention provides a computer program product which, when running on a computer, causes the computer to execute the intelligent traffic signal control method based on Agent-ARIMA as described above.

[0030] Compared with the prior art, the present invention has the following technical effects:

[0031] (1) Real-time acquisition of road section information: By calling the map API to obtain road section information in real time, it can adapt to the current traffic flow, improve the real-time and accuracy of signal light time calculation, and provide a high-quality input data basis for subsequent calculations.

[0032] (2) Improvement in short-term prediction accuracy: The ARIMA model can provide high-precision predictions for short-term traffic flow. Using its results as weights and inputting them into the LLM model helps the model better handle short-term changes in traffic flow, especially playing a role in the rapid traffic flow fluctuations during the morning and evening rush hours.

[0033] (3) Combination of global and local: The LLM model is good at analyzing complex traffic patterns from a global perspective, while the ARIMA model performs well in local short-term predictions. Using the output of the ARIMA model as weights allows the LLM model to fully consider local short-term changes during global predictions, improving the overall model's response speed and flexibility.

[0034] (4) Improvement in model stability and robustness: The ARIMA model can effectively reduce the prediction uncertainty caused by short-term fluctuations in traffic flow, endowing the LLM model with more stable input weights and enhancing the model's robustness in the face of sudden traffic conditions.

[0035] (5) Resource optimization and improvement in computational efficiency: Since the computational complexity of the ARIMA algorithm is relatively low, by integrating its results as weights into the LLM model, the computational burden of the LLM model for short-term fluctuations can be reduced, thereby optimizing the overall model's resource usage and improving computational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] By reading the detailed description of the non-limiting embodiments with reference to the following drawings, the present invention and its features and advantages will become more obvious.

[0037] Figure 1 It is a flowchart of the intelligent traffic signal control method in an embodiment of the present invention;

[0038] Figure 2 It is a schematic diagram of a single intersection model in an embodiment of the present invention;

[0039] Figure 3 It is a flowchart of obtaining the calculation results of multiple intersections through a single intersection in an embodiment of the present invention. Detailed implementation manners

[0040] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but it is not a limitation of the present invention.

[0041] In the following detailed description, many specific details are set forth to provide a more thorough understanding of the present invention. However, it is obvious to those skilled in the art that well-known algorithms and models are not shown in detail to avoid obscuring the gist of the present invention.

[0042] In addition, for the execution order of actions, steps, etc. in the devices and methods shown in the claims, the description and the drawings, as long as there is no specific limitation on the order, and as long as the output of the previous process is not used in the subsequent process, it can be implemented in any order.

[0043] Embodiment

[0044] Refer to Figure 1 , this embodiment provides an intelligent traffic signal control method based on Agent-ARIMA, including the following steps:

[0045] S1. Call the map API to obtain the congestion distance, average driving speed, traffic flow of the left-turn lane and the straight-through lane of each road section at the crossroads in real time.

[0046] S2. Calculate the maximum time for all vehicles on two intersecting roads to pass through the traffic lights . Set the shortest green light duration , which can enable pedestrians to pass through the zebra crossing. This method comprehensively considers the shortest passing time of pedestrians , the longest tolerance time of drivers , and mainly distributes time according to the proportion of the longest dredging time of this road section. Set the shortest green light duration to ensure that pedestrians can pass through the zebra crossing.

[0047] S3. Perform the green light time allocation at the intersection. To reduce the waiting time of pedestrians and balance the vehicle traffic, the length of the green light cycle is determined according to the required green light time in each direction and , and the formula is . The lane directions of a single intersection are as Figure 2 shown. When there are too many vehicles, to shorten the waiting time of drivers, according to the required green light time of road bd and the required green light time of road ac , the traffic light cycle length is determined to be .

[0048] S4. Left-turn lane and straight-through lane time allocation: Allocate green light time for the left-turn lane and straight-through lane at the intersection. When the required green light time for the x direction is , and the required green light time for the y direction is , ensure that and . If this condition is not met for any direction, adjust the green light duration for the next cycle. If one side is less than , then set the green light time for this side to in the next cycle. However, the longest duration of the determined traffic light cycle must not be greater than the longest cycle .

[0049] S5. Dynamic adjustment of traffic cycle: Calculate the actual green light time for each direction in the next cycle based on the traffic flow ratio in each direction. When the required green light duration is and it can ensure that pedestrians can cross the intersection, that is, and , if the determined traffic light cycle duration does not exceed the set longest cycle , set the required green light durations for the ab direction and cd direction as the actual green light durations for the ab direction and cd direction in the next cycle, that is, , ; if the determined traffic light cycle duration exceeds the set longest cycle , according to ratio-based timing, allocate the longest cycle based on the proportion that the required green light duration occupies in its determined traffic light cycle duration , that is, according to traffic demand, the green light duration for the ab direction and the green light duration for the cd direction are respectively: .

[0050] S6. Perform weight offset based on a single intersection to obtain multi-intersection time. The specific step flowchart is as shown in Figure 3 . This step requires obtaining road section information, including: congestion distance, average vehicle speed, congestion status, etc. The required green light time for the ac road, and the required green light time for the bd road to obtain the distance matrix between each intersection. Calculate the new green light time allocation ratio based on various factors affecting road section congestion and target traffic flow. Adjust the green light time ratio by calculating the weight range . Further, the distance between intersections can calculate the time for traffic flow to reach the next intersection based on the current average vehicle speed of this road section Calculate the time it takes for vehicles to reach the current intersection from the upstream intersection , where S is the distance between two adjacent upstream and downstream intersections, and v is the vehicle flow speed. Save the result as an array; calculate the current time , and determine whether it is less than the time required to reach the intersection plus the time to obtain data , that is seconds. If it is greater, then keep this time. If the condition is not met, set the time value at the corresponding position in the weight array to 0, and at the same time, re-fill the vehicle flow information at the non-zero positions in the weight array according to the real-time traffic flow data. Take the green light time of the previous period on the same road as the weight generation array and store it. On this basis, check whether there is a value of 0 in the weight array. If there is, the offset weight needs to be recalculated. At the same time, the system will check whether it is less than the maximum green light time , and whether it meets or . If these conditions are met, directly output the and of the current cycle; if not, enter the calculation link of the offset weight. The time ratio of transferring to is called the offset weight, and the weight calculation formula is: , make duration transfer adjustments to and . Obtain the right shift ratio by looking up the table, and make duration transfer adjustments to and accordingly. If the adjusted and are less than the minimum green light duration , then execute the offset callback mechanism to adjust the green light duration to be greater than or equal to . Finally, the system outputs the dynamically adjusted and of the current cycle, providing an optimized basis for the green light time allocation of the next cycle

[0051] S7. According to the ARIMA algorithm (AutoRegressive Integrated Moving Average), use the time series to fit the ARIMA model, and predict the future time period based on the model to obtain the predicted value .

[0052] ARIMA (Autoregressive Integrated Moving Average Model) is a classic algorithm for time series prediction. It combines the autoregressive model (AR), the moving average model (MA), and the differencing process, and is applicable to the prediction of non-stationary time series. The ARIMA model transforms a non-stationary time series into a stationary time series through differencing operations, and then uses the AR and MA models for prediction.

[0053] First, define the ARIMA model. The form of the ARIMA model is ARIMA(p, d, q), where p is the order of the autoregressive part (AR part), d is the order of differencing, representing the number of differencing times required for the data to be stationary (I part), and q is the order of the moving average part (MA part). The ARIMA model can be expressed as: ;

[0054] where, is the constant term, is the coefficient of the autoregressive term, is the coefficient of the moving average term, is the error term, which is usually regarded as white noise.

[0055] S8. Determine whether the traffic conditions at the intersection can be correctly predicted. Intersections with a mean absolute percentage error less than 10% are considered to be correctly predicted, and such intersections are called "simple intersections". If the error is greater than 10%, such intersections are called "complex intersections". The formula for calculating the mean absolute percentage error is: ;

[0056] where, is the number of groups of prediction data, is the predicted result value such as traffic flow, is the actual measured value.

[0057] If the intersection is a complex intersection, the signal control time series predicted for the intersection based on the ARIMA model is used as input data and input into the LLM-Agent signal control agent. Specifically, the output of the ARIMA model is the predicted signal light durations arranged in chronological order (such as the durations of red, green, and yellow lights), and each time point contains the corresponding signal light duration information. After preprocessing and standardization, this time series data is input into the time series analysis sub-module of the LLM-Agent in the form of embedding vectors or direct numerical values to achieve decision optimization and dynamic adjustment of signal control by the agent and predict the signal light time. Among them, the construction steps of the LLM model fine-tuned and trained based on traffic data include, first, constructing traffic state observation features and collecting traffic state observations at traffic intersections; second, constructing intelligent agent prompts enhanced with common sense knowledge to form a prompt integrating common sense knowledge for guiding the LLM to infer the optimal traffic signal configuration for the next time slice; finally, analyzing, reasoning, and making decisions by the agent. The LLM uses the constructed prompt for the analysis, reasoning, and decision-making process and then makes a decision.

[0058] S9. Obtain the latest road conditions after a fixed time interval, and update the times of the two types of intersections, namely "simple intersections" and "complex intersections", according to the latest road conditions. For simple intersections, only steps S1 to S7 need to be repeated, and for complex intersections, steps S1 to S9 need to be repeated additionally.

[0059] A control system implementing the intelligent traffic signal control method based on Agent-ARIMA as described above includes:

[0060] A road section information collection module that collects the congestion distance, average driving speed, traffic flow of the left-turn lane and the straight-through lane at each road section of the crossroads;

[0061] An intersection signal light time allocation module that calculates the maximum time for all vehicles on two intersecting roads to pass through the traffic lights , sets the shortest green light duration , conducts intersection green light time allocation, left-turn lane and straight-through lane time allocation, and dynamic adjustment of the traffic cycle; performs weight offset on the basis of a single intersection to conduct multi-intersection time allocation;

[0062] An ARIMA model that, according to the ARIMA algorithm, uses a time series to fit the ARIMA model and predicts future time periods based on the ARIMA model to obtain predicted values ;

[0063] LLM-Agent signal light control agent, a large language model fine-tuned and trained based on traffic data; for complex intersections, the signal light control time series predicted based on the ARIMA model is used as the input data of the LLM-Agent signal light control agent to predict the signal light time.

[0064] The above-mentioned intelligent traffic signal control method based on Agent-ARIMA can be embodied in the form of a computer program product or a software functional unit. If the above-mentioned intelligent traffic signal control method based on Agent-ARIMA is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Therefore, the essence of this technical solution, or the part that contributes to the prior art, or a part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable an electronic system (which can be a personal computer, a server, or a network system, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, external hard drives, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0065] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with this embodiment can be implemented by electronic hardware or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0066] In summary, the present invention provides an intelligent traffic signal control method, system, and program product based on Agent-ARIMA. The intelligent traffic signal control method includes steps such as calculating the shortest green light time for each intersection direction, allocating the green light time for each direction, allocating the green light time for left-turn and straight lanes, adjusting the traffic cycle, calculating the multi-intersection time weight offset, using the ARIMA model to predict future traffic flow, and for complex intersections, inputting the time predicted by the ARIMA algorithm into the LLM-Agent intelligent body for time prediction. The present invention is applicable to different city scales. By accurately predicting and analyzing traffic flow and dynamically adjusting the time allocation of traffic lights, it can not only optimize traffic congestion in large cities but also improve traffic efficiency in small cities.

[0067] Those skilled in the art should understand that those skilled in the art can implement variations in combination with the prior art and the above embodiments, which will not be elaborated here. Such variations do not affect the essence of the present invention and will not be elaborated here.

[0068] The preferred embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the above specific embodiments, and the systems and structures not described in detail should be understood to be implemented in a common manner in the art; any person skilled in the art can make many possible changes and modifications to the technical solution of the present invention, or modify it into an equivalent embodiment with equivalent changes, without departing from the scope of the technical solution of the present invention, which does not affect the essence of the present invention. Therefore, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. An intelligent traffic signal control method based on Agent-ARIMA, characterized in that: The following steps are involved: Step S1, calling the map API to obtain the congestion distance, average driving speed, and traffic volume of the left turn lane and the straight lane of each section of the intersection in real time; Step S2: Calculate the maximum time for all vehicles on the two intersections to pass the traffic lights , set the minimum green light duration , Enable pedestrians to cross the zebra crossing; Step S3: Green light time allocation at intersection: green light cycle length According to the green light time required for each direction and OK, the formula is ; Step S4, time allocation between left turn lane and straight lane: When the required green light time in the x direction is , the green light time required in the y direction is When making sure and ; If this condition is not met in any direction, the green light duration of the next cycle is adjusted; If one side is smaller than , then in the next cycle, the green light time on this side is set to ; Step S5, dynamic adjustment of traffic cycle: calculate the actual green light time of each direction in the next cycle according to the traffic volume ratio in each direction; the green light time of directions a and b And the green light duration in the cd direction They are: ； Step S6, weight offset calculation of multiple intersection times: obtain road section information, required green light time of AC road , BD Road required green light time Get the distance matrix between each intersection; calculate the new green light time allocation ratio based on multiple factors including traffic congestion and traffic volume in both directions; calculate the weight range , adjust the green light time ratio; Step S7: According to the ARIMA algorithm, use the time series To fit the ARIMA model, make predictions for future time periods based on the model to get the predicted values ; The ARIMA model can be expressed as: in, is a constant term, is the coefficient of the autoregressive term, is the coefficient of the sliding average term, is the error term, usually regarded as white noise, p‌: the order of the autoregressive model, which indicates the linear relationship between the current value and the past p values; q‌: the order of the moving average model, which indicates the linear relationship between the current value and the past q prediction errors; Step S8, judging whether the traffic situation at the intersection can be correctly predicted, the intersection with a mean absolute percentage error of less than 10% is considered to be correctly predicted, and such intersection is called a "simple intersection"; if the error is greater than 10%, such intersection is called a "complex intersection"; if the intersection is a complex intersection, the traffic light control time series predicted by the ARIMA model for the intersection is used as input data and input into the LLM-Agent traffic light control agent to predict the traffic light time; the LLM-Agent traffic light control agent is a large language model fine-tuned and trained based on traffic data; Step S9, obtaining the latest road conditions after a fixed time interval, and updating the intersection times of "simple intersection" and "complex intersection" according to the latest road conditions.

2. The intelligent traffic signal control method based on Agent-ARIMA according to claim 1, characterized in that: In step S6, the road section information includes but is not limited to congestion distance, average vehicle speed, and congestion status.

3. The intelligent traffic signal control method based on Agent-ARIMA according to claim 1, characterized in that: In step S6, the distance between intersections can be used to calculate the time it takes for traffic flow to reach the next intersection based on the current average speed of the road section. Calculate the time it takes for the upstream intersection to reach the current intersection , where S is the distance between two adjacent upstream and downstream intersections, v is the traffic speed, and is saved as an array; calculate the current time , determine whether it is less than the time required to arrive plus the time to obtain data ,Right now Seconds, if greater than, keep the time; If it is insufficient, take the green light time of the previous period on the same road as the weight to generate an array and store it; Transfer to The time proportion is called the offset weight, and the weight calculation formula is: ,right and Make time shift adjustments.

4. The intelligent traffic signal control method based on Agent-ARIMA according to claim 1 is characterized in that: In step S8, the output of the ARIMA model is the predicted traffic light duration arranged in chronological order, and each time point contains the corresponding traffic light duration information; the predicted traffic light duration arranged in chronological order is standardized through preprocessing and input into the time series analysis submodule of the LLM-Agent traffic light control agent in the form of embedded vectors or direct numerical values.

5. The intelligent traffic signal control method based on Agent-ARIMA according to claim 1 or 4, characterized in that: In step S8, the steps of constructing the LLM model after fine-tuning the training based on traffic data include: first, constructing traffic status observation features and collecting traffic status observation data at traffic intersections; second, constructing a model prompt enhanced with common sense knowledge to form a prompt that integrates common sense knowledge to guide the LLM model to infer the optimal traffic light configuration for the next time slice; finally, the LLM model uses the constructed prompt to perform an analysis and reasoning decision-making process, and then makes a decision.

6. The intelligent traffic signal control method based on Agent-ARIMA according to claim 1, characterized in that: In step S9, for a "simple intersection", repeat steps S1 to S7, and for a "complex intersection", repeat steps S1 to S9.

7. An intelligent traffic signal control system based on Agent-ARIMA, characterized in that: The method for implementing the intelligent traffic signal control method based on Agent-ARIMA as claimed in any one of claims 1 to 6 comprises: The road section information collection module collects the congestion distance, average driving speed, and traffic volume of the left-turn lane and the through lane at each section of the intersection; The intersection signal light time allocation module calculates the maximum time for all vehicles on two intersections to pass the traffic lights , set the minimum green light duration , allocate green light time at intersections, allocate time for left-turn lanes and through lanes, and dynamically adjust traffic cycles; perform weight shifting based on a single intersection and allocate time for multiple intersections; ARIMA model, based on the ARIMA algorithm, uses time series To fit the ARIMA model, predict the future time period based on the ARIMA model to get the predicted value ; The LLM-Agent signal light control agent fine-tunes the trained large language model based on traffic data; for complex intersections, the signal light control time series predicted by the ARIMA model is used as the input data of the LLM-Agent signal light control agent to predict the signal light time.

8. A computer program product, characterized in that When the computer program product is run on a computer, the computer is enabled to execute the intelligent traffic signal control method based on Agent-ARIMA as claimed in any one of claims 1 to 6.

Images