An AI-based urban traffic flow prediction and dynamic signal optimization method and system

Abstract

The application discloses an AI-based urban traffic flow prediction and dynamic signal optimization method and system, and relates to the technical field of traffic management.The method comprises the following steps: acquiring real-time traffic data, historical flow data and external environment data of a target area to obtain multi-source data, and constructing a space-time matrix corresponding to the target area according to the multi-source data; learning the space-time matrix by using a preset space-time fusion model, and outputting a prediction result of traffic flow in a future preset time period by the learned space-time fusion model; wherein the space-time fusion model is a joint architecture of a graph convolution network and a Transformer network; and based on the prediction result, dynamically adjusting traffic signal control parameters in the target area by using a multi-agent learning algorithm.The application forms a closed-loop regulation and control of "data perception-prediction modeling-autonomous decision-making", thereby breaking through the limitations of the traditional method of space feature modeling staticization and time correlation analysis fragmentation, and effectively solving the regional imbalance problem caused by single-point optimization.

Description

Technical Field

[0001] This invention relates to the field of traffic management technology, specifically to an AI-based method and system for urban traffic flow prediction and dynamic signal optimization. Background Technology

[0002] In the rapid development of smart cities, traditional traffic management models are struggling to cope with the systemic congestion caused by the surge in motor vehicle ownership. Existing traffic control systems generally adopt a "detection-rule response" mechanism, whose limitations, such as limited data dimensions and delayed decision-making, have become key bottlenecks restricting the modernization of urban governance.

[0003] In existing technologies, traffic flow prediction models often process spatial topology and time-series features independently, leading to two core problems: First, static road network modeling cannot reflect the dynamic decline in actual traffic efficiency (such as cascading congestion at related intersections caused by reduced vehicle speeds in rainy weather); second, time-series models such as LSTM have long response delays to sudden events, making it difficult to meet real-time control requirements. Furthermore, traditional signal optimization uses a centralized computing architecture, and when the number of intersections is large, the time required for collaborative decision-making increases exponentially, resulting in a high regional imbalance rate. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide an AI-based method and system for urban traffic flow prediction and dynamic signal optimization, thereby solving the problems in existing technologies.

[0005] The first aspect of this invention is to provide an AI-based method for urban traffic flow prediction and dynamic signal optimization, the method comprising:

[0006] Real-time traffic data, historical flow data and external environment data of the target area are acquired to obtain multi-source data, and a spatiotemporal matrix corresponding to the target area is constructed based on the multi-source data.

[0007] The spatiotemporal matrix is ​​learned using a pre-defined spatiotemporal fusion model, and the resulting spatiotemporal fusion model outputs a prediction of traffic flow within a pre-defined time period. The spatiotemporal fusion model is a joint architecture of graph convolutional network and Transformer network.

[0008] Based on the prediction results, the traffic signal control parameters within the target area are dynamically adjusted using a multi-agent learning algorithm.

[0009] According to one aspect of the above technical solution, the step of acquiring real-time traffic data, historical flow data, and external environment data of the target area to obtain multi-source data, and constructing a spatiotemporal matrix corresponding to the target area based on the multi-source data, includes:

[0010] Real-time traffic data, historical flow data, and external environment data of the target area are acquired, and the real-time traffic data, historical flow data, and external environment data are fused to obtain multi-source data;

[0011] Based on the multi-source data, the road network topology and time series data were extracted respectively;

[0012] The intersections within the target area are modeled as graph nodes using the road network topology, spatial weights between adjacent intersections are extracted, and the time series data is periodically decomposed to extract periodic features from the time series data.

[0013] Based on the spatial weights and the periodic features, a spatiotemporal matrix corresponding to the target region is constructed.

[0014] According to one aspect of the above technical solution, the step of learning the spatiotemporal matrix using a preset spatiotemporal fusion model and outputting a prediction result of traffic flow within a preset time period using the learned spatiotemporal fusion model includes:

[0015] The spatial weights and periodic features in the spatiotemporal matrix are learned using a pre-defined spatiotemporal fusion model to obtain the learned spatiotemporal fusion model.

[0016] The spatiotemporal fusion model outputs the predicted traffic flow for a future preset time period.

[0017] According to one aspect of the above technical solution, the step of dynamically adjusting the traffic signal control parameters within the target area based on the prediction results using a multi-agent learning algorithm includes:

[0018] Based on the prediction results, the node features of each traffic control node in the target area are output through a multi-agent learning algorithm;

[0019] The system controls each traffic control node to perform centralized training and decentralized execution, outputs the phase / green light duration of each traffic control node, and evaluates the global value score of each traffic control node.

[0020] Based on the global value score of the traffic control node, the phase / green light duration corresponding to the traffic control node is adjusted according to the time period characteristics.

[0021] According to one aspect of the above technical solution, the step of adjusting the phase / green light duration corresponding to the traffic control node according to the global value score of the traffic control node and the time period characteristics includes:

[0022] Every first preset time interval, the traffic prediction results corresponding to the target area are received from the cloud, and the global value score of the traffic control node is updated according to the traffic prediction results.

[0023] The control edge regenerates the signal control scheme for each traffic control node every second preset time interval. The signal control scheme includes the phase selection probability and the green light duration correction value.

[0024] According to one aspect of the above technical solution, the control edge regenerates the signal control scheme for each traffic control node every second preset time interval. The signal control scheme includes steps for phase selection probability and green light duration correction, comprising:

[0025] The control edge runs the Actor network every second preset time interval, and regenerates the signal control scheme for each traffic control node through the Actor network;

[0026] The Actor network takes the local state and neighbor states within the target area as input and outputs an action probability distribution, which includes the phase selection probability and the green light duration correction value.

[0027] A second aspect of the present invention is to provide an AI-based urban traffic flow prediction and dynamic signal optimization system, applied to the method described in the above-mentioned technical solution, the system comprising:

[0028] The data acquisition module is used to acquire real-time traffic data, historical flow data and external environment data of the target area to obtain multi-source data, and construct a spatiotemporal matrix corresponding to the target area based on the multi-source data;

[0029] The prediction output module is used to learn the spatiotemporal matrix using a preset spatiotemporal fusion model, and output the prediction result of traffic flow within a preset time period through the learned spatiotemporal fusion model; wherein, the spatiotemporal fusion model is a joint architecture of graph convolutional network and Transformer network;

[0030] The dynamic optimization module is used to dynamically adjust the traffic signal control parameters within the target area based on the prediction results using a multi-agent learning algorithm.

[0031] According to one aspect of the above technical solution, the data acquisition module is used for:

[0032] Real-time traffic data, historical flow data, and external environment data of the target area are acquired, and the real-time traffic data, historical flow data, and external environment data are fused to obtain multi-source data;

[0033] Based on the multi-source data, the road network topology and time series data were extracted respectively;

[0034] The intersections within the target area are modeled as graph nodes using the road network topology, spatial weights between adjacent intersections are extracted, and the time series data is periodically decomposed to extract periodic features from the time series data.

[0035] Based on the spatial weights and the periodic features, a spatiotemporal matrix corresponding to the target region is constructed.

[0036] A third aspect of the present invention is to provide a readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the methods described in the above-described technical solutions.

[0037] A fourth aspect of the present invention is to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the methods described in the above-described technical solutions.

[0038] Compared with existing technologies, the advantages of using the AI-based urban traffic flow prediction and dynamic signal optimization method and system shown in this embodiment are as follows:

[0039] This invention combines graph convolutional networks to extract dynamic spatial dependencies in road networks with Transformers to analyze multi-scale temporal features, achieving accurate traffic flow prediction. Based on this, it employs a multi-agent collaborative control architecture with centralized training and distributed execution, enabling each intersection to optimize signal parameters in real time based on local state and global reward signals. This forms a closed-loop control system of "data perception - predictive modeling - autonomous decision-making," overcoming the limitations of traditional methods such as static spatial feature modeling and fragmented temporal correlation analysis. It effectively solves the regional imbalance problem caused by single-point optimization, while reducing decision latency through edge computing, achieving a triple technical effect of improved prediction accuracy, enhanced real-time control, and optimized regional traffic efficiency, systematically improving urban traffic congestion. Attached Figure Description

[0040] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0041] Figure 1 This is a flowchart illustrating an AI-based urban traffic flow prediction and dynamic signal optimization method according to an embodiment of the present invention.

[0042] Figure 2 This is a structural block diagram of an AI-based urban traffic flow prediction and dynamic signal optimization system according to an embodiment of the present invention. Detailed Implementation

[0043] To make the objectives, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Several embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of the present invention will be more thorough and complete.

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0045] Example 1

[0046] Please see Figure 1 The first aspect of the present invention provides an AI-based method for urban traffic flow prediction and dynamic signal optimization. The method shown in this embodiment includes steps S10-S30:

[0047] Step S10: Obtain real-time traffic data, historical flow data and external environment data of the target area to obtain multi-source data, and construct a spatiotemporal matrix corresponding to the target area based on the multi-source data.

[0048] In this embodiment, the target area is the area where traffic flow prediction and dynamic signal optimization are required. It is usually a key area in the city, which typically contains main urban roads and various supporting facilities. Traffic flow changes dynamically over time. For example, the areas where shopping malls and schools are located are the target areas shown in this embodiment.

[0049] The multi-source data includes:

[0050] Real-time traffic data: flow and speed information from cameras, geomagnetic sensors, and vehicle GPS;

[0051] External environmental data: weather conditions, holiday signs, emergencies, and road construction plans;

[0052] Historical traffic data: Intersection traffic volume and signal phase records stored in the form of time segments.

[0053] Specifically, in this embodiment, real-time traffic data, historical flow data and external environment data of the target area are acquired, and the real-time traffic data, historical flow data and external environment data are fused to obtain multi-source data. Then, a spatiotemporal matrix corresponding to the target area is constructed based on the multi-source data. The constructed spatiotemporal matrix is ​​used to express the correspondence between traffic flow and time and space in the target area.

[0054] More specifically, the real-time traffic data of the target area is acquired based on cameras, geomagnetic sensors, and vehicle GPS. It typically includes the real-time average vehicle speed and the average waiting time at intersections within the target area. The acquired historical traffic data includes the average vehicle speed and the average waiting time at intersections at previous times. The external environment data mainly includes the supporting facilities of the target area and records the spatial information within the target area. The above data are then fused to obtain multi-source data, and a spatiotemporal matrix is ​​constructed based on this to express the temporal and spatial correspondence of traffic flow within the target area.

[0055] Step S20: The spatiotemporal matrix is ​​learned using a preset spatiotemporal fusion model, and the predicted traffic flow within a preset time period is output through the learned spatiotemporal fusion model.

[0056] The spatiotemporal fusion model is a joint architecture of graph convolutional networks and Transformer networks.

[0057] Specifically, after acquiring real-time traffic data, historical flow data, and external environment data of the target area to obtain multi-source data and construct a spatiotemporal matrix, the pre-constructed spatiotemporal matrix will be learned using a pre-set spatiotemporal fusion model. Through learning, the spatiotemporal fusion model will output the predicted traffic flow of the target area within a future preset time period.

[0058] In this embodiment, the step of learning the spatiotemporal matrix using a preset spatiotemporal fusion model and outputting the predicted traffic flow for a preset time period using the learned spatiotemporal fusion model includes:

[0059] The spatial weights and periodic features in the spatiotemporal matrix are learned using a pre-defined spatiotemporal fusion model to obtain the learned spatiotemporal fusion model.

[0060] The spatiotemporal fusion model outputs the predicted traffic flow for a future preset time period.

[0061] More specifically, in this embodiment, a preset spatiotemporal fusion model is used to learn the spatial weights and periodic features in the spatiotemporal matrix to obtain the learned spatiotemporal fusion model. This enables the learned spatiotemporal fusion model to more accurately extract features based on multi-source data and thus predict traffic flow.

[0062] Step S30: Based on the prediction results, dynamically adjust the traffic signal control parameters within the target area using a multi-agent learning algorithm.

[0063] After predicting the traffic flow structure within the target area using the spatiotemporal fusion model learned from the model, a multi-agent learning algorithm will be used to dynamically adjust the traffic control signal parameters of multiple traffic control nodes within the target area. This includes adjusting the red and green light durations of traffic lights and the turning signals of variable lanes in one or more traffic control nodes.

[0064] Specifically, when dynamically adjusting traffic control signal parameters using a multi-agent learning algorithm, traffic control nodes are treated as agents. First, based on traffic flow prediction results, the node characteristics of each traffic control node within the target area are output. Each traffic control node is then trained in a centralized manner and executed in a decentralized manner. The phase / green light duration of each traffic control node is output. The global value score of each traffic control node is evaluated based on its phase / green light duration. This typically depends on the location of each traffic control node and the supporting facilities in the surrounding area. Then, based on the global value score of each traffic control node, the global value scores of all traffic control nodes within the target area are sorted. The phase / green light duration of each traffic control node is then dynamically adjusted according to time period characteristics.

[0065] More specifically, the cloud-based backend server re-predicts the traffic flow forecast for the target area at the current time node based on previously acquired multi-source data every first preset time interval, such as every 5-10 minutes. It receives the traffic flow forecast results corresponding to the target area synchronously sent by the cloud-based backend server in real time, and then dynamically updates the global value score of each traffic control node based on the current traffic flow forecast results. After updating the global value score of each traffic control node, the control edge regenerates the signal control scheme for each traffic control node every second preset time interval according to the updated global value score. The second preset time interval is the same as the first preset time interval. The signal control scheme includes the phase selection probability and the green light duration correction value. Then, the signal control scheme is executed to achieve dynamic adjustment of the traffic signal.

[0066] The process of regenerating the signal control scheme for each traffic control node at the control edge every second preset time interval specifically includes running the Actor network at the control edge every second preset time interval. The Actor network regenerates the signal control scheme for each traffic control node through the operation of the Actor network. The Actor network iterates through the local state of a traffic control node within the target area and the neighbor states of other traffic control nodes within the target area, outputting an action probability distribution. This action probability distribution includes a phase selection probability and a green light duration correction value. Specifically, the green light duration correction value for a traffic control node is calculated based on its phase selection probability. For example, if the green light duration of the traffic control node's signal light was previously 50 seconds, after querying the neighbor states of other traffic control nodes, it can be determined that the traffic flow in the target area has significantly increased or decreased. Based on the association between the traffic control node and other traffic control nodes, the green light duration of the traffic control node's signal light is corrected, for example, by adjusting the green light duration to 80 seconds or 30 seconds, so that the traffic light corresponding to the traffic control node has a longer or shorter green light duration.

[0067] For example, when a traffic accident at an intersection causes congestion, the method shown in this embodiment quickly acquires current multi-source data and constructs a spatiotemporal matrix. Then, it uses a spatiotemporal fusion model to predict the traffic flow in the target area. Finally, it combines a multi-agent learning algorithm to dynamically adjust the phase selection probability of upstream and downstream traffic control nodes and the green light duration of traffic lights in the target area. By responding quickly and formulating the optimal signal control scheme, and combining it with the corresponding traffic condition feedback, it can effectively alleviate traffic congestion and prevent it from worsening.

[0068] Compared with existing technologies, the AI-based urban traffic flow prediction and dynamic signal optimization method shown in this embodiment has the following advantages:

[0069] This embodiment combines graph convolutional networks to extract dynamic spatial dependencies of the road network with Transformer to parse multi-scale temporal features, achieving accurate traffic flow prediction. Based on this, a multi-agent collaborative control architecture with centralized training and distributed execution is adopted, enabling each intersection to optimize signal parameters in real time based on local state and global reward signals. This forms a closed-loop control system of "data perception - predictive modeling - autonomous decision-making," overcoming the limitations of traditional methods such as static spatial feature modeling and fragmented temporal correlation analysis. It effectively solves the regional imbalance problem caused by single-point optimization, while reducing decision latency through edge computing, achieving a triple technical effect of improved prediction accuracy, enhanced real-time control, and optimized regional traffic efficiency, systematically improving urban traffic congestion.

[0070] Example 2

[0071] The second embodiment of the present invention also provides an AI-based method for urban traffic flow prediction and dynamic signal optimization. The method shown in this embodiment is basically similar to the method shown in the first embodiment, except that:

[0072] In this embodiment, the steps of acquiring real-time traffic data, historical flow data, and external environment data of the target area to obtain multi-source data, and constructing a spatiotemporal matrix corresponding to the target area based on the multi-source data, include:

[0073] Real-time traffic data, historical flow data, and external environment data of the target area are acquired, and the real-time traffic data, historical flow data, and external environment data are fused to obtain multi-source data;

[0074] Based on the multi-source data, the road network topology and time series data were extracted respectively;

[0075] The intersections within the target area are modeled as graph nodes using the road network topology, spatial weights between adjacent intersections are extracted, and the time series data is periodically decomposed to extract periodic features from the time series data.

[0076] Based on the spatial weights and the periodic features, a spatiotemporal matrix corresponding to the target region is constructed.

[0077] Specifically, after acquiring and fusing real-time traffic data, historical flow data, and external environment data of the target area to obtain multi-source data, the road network topology and corresponding traffic flow time series data of the target area are extracted based on the multi-source data. All intersections in the target area are modeled as multiple graph nodes through the road network topology, and the spatial weights between adjacent intersections are extracted. Then, the time series data is periodically decomposed to extract the periodic features in the time series data. Finally, based on the spatial weight of each intersection and the periodic features of the corresponding traffic flow, a spatiotemporal matrix corresponding to the target area is constructed. The spatiotemporal matrix is ​​used to express the dual correspondence between traffic flow and time and space of each intersection in the target area.

[0078] More specifically, before each adjustment to the signal control scheme, it is necessary to acquire real-time multi-source data within the target area and then construct the corresponding spatiotemporal matrix. This allows for traffic flow prediction through a pre-built spatiotemporal fusion model, thereby enabling dynamic adjustments to the signal control scheme.

[0079] Example 3

[0080] Please see Figure 2 The third embodiment of the present invention provides an AI-based urban traffic flow prediction and dynamic signal optimization system, applied to the method described in any of the above embodiments, the system comprising:

[0081] Data acquisition module 10 is used to acquire real-time traffic data, historical flow data and external environment data of the target area to obtain multi-source data, and construct a spatiotemporal matrix corresponding to the target area based on the multi-source data;

[0082] The prediction output module 20 is used to learn the spatiotemporal matrix using a preset spatiotemporal fusion model, and output the prediction result of traffic flow within a preset time period through the learned spatiotemporal fusion model; wherein, the spatiotemporal fusion model is a joint architecture of graph convolutional network and Transformer network;

[0083] The dynamic optimization module 30 is used to dynamically adjust the traffic signal control parameters within the target area based on the prediction results using a multi-agent learning algorithm.

[0084] The data acquisition module 10 is used for:

[0085] Real-time traffic data, historical flow data, and external environment data of the target area are acquired, and the real-time traffic data, historical flow data, and external environment data are fused to obtain multi-source data;

[0086] Based on the multi-source data, the road network topology and time series data were extracted respectively;

[0087] The intersections within the target area are modeled as graph nodes using the road network topology, spatial weights between adjacent intersections are extracted, and the time series data is periodically decomposed to extract periodic features from the time series data.

[0088] Based on the spatial weights and the periodic features, a spatiotemporal matrix corresponding to the target region is constructed.

[0089] Compared with existing technologies, the advantages of using the AI-based urban traffic flow prediction and dynamic signal optimization system shown in this embodiment are as follows:

[0090] This embodiment combines graph convolutional networks to extract dynamic spatial dependencies of the road network with Transformer to parse multi-scale temporal features, achieving accurate traffic flow prediction. Based on this, a multi-agent collaborative control architecture with centralized training and distributed execution is adopted, enabling each intersection to optimize signal parameters in real time based on local state and global reward signals. This forms a closed-loop control system of "data perception - predictive modeling - autonomous decision-making," overcoming the limitations of traditional methods such as static spatial feature modeling and fragmented temporal correlation analysis. It effectively solves the regional imbalance problem caused by single-point optimization, while reducing decision latency through edge computing, achieving a triple technical effect of improved prediction accuracy, enhanced real-time control, and optimized regional traffic efficiency, systematically improving urban traffic congestion.

[0091] Example 4

[0092] A fourth embodiment of the present invention provides a readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any of the above embodiments.

[0093] Example 5

[0094] A fifth embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described in any of the above embodiments.

[0095] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0096] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. An AI-based method for urban traffic flow prediction and dynamic signal optimization, characterized in that, The method includes: Real-time traffic data, historical flow data and external environment data of the target area are acquired to obtain multi-source data, and a spatiotemporal matrix corresponding to the target area is constructed based on the multi-source data. The spatiotemporal matrix is ​​learned using a pre-defined spatiotemporal fusion model, and the resulting spatiotemporal fusion model outputs a prediction of traffic flow within a pre-defined time period. The spatiotemporal fusion model is a joint architecture of graph convolutional network and Transformer network. Based on the prediction results, the traffic signal control parameters within the target area are dynamically adjusted using a multi-agent learning algorithm. The steps of acquiring real-time traffic data, historical flow data, and external environment data of the target area to obtain multi-source data, and constructing a spatiotemporal matrix corresponding to the target area based on the multi-source data, include: Real-time traffic data, historical flow data, and external environment data of the target area are acquired, and the real-time traffic data, historical flow data, and external environment data are fused to obtain multi-source data; Based on the multi-source data, the road network topology and time series data were extracted respectively; The intersections within the target area are modeled as graph nodes using the road network topology, spatial weights between adjacent intersections are extracted, and the time series data is periodically decomposed to extract periodic features from the time series data. Based on the spatial weights and the periodic features, a spatiotemporal matrix corresponding to the target region is constructed; Based on the prediction results, the step of dynamically adjusting the traffic signal control parameters within the target area using a multi-agent learning algorithm includes: Based on the prediction results, the node features of each traffic control node in the target area are output through a multi-agent learning algorithm; The system controls each traffic control node to perform centralized training and decentralized execution, outputs the phase / green light duration of each traffic control node, and evaluates the global value score of each traffic control node. Based on the global value score of the traffic control node, the phase / green light duration corresponding to the traffic control node is adjusted according to the time period characteristics; The control edge regenerates the signal control scheme for each traffic control node every second preset time interval. The signal control scheme includes steps for phase selection probability and green light duration correction, including: The control edge runs the Actor network every second preset time interval, and regenerates the signal control scheme for each traffic control node through the Actor network; The Actor network takes the local state and neighbor states within the target area as input and outputs an action probability distribution, which includes the phase selection probability and the green light duration correction value.

2. The AI-based urban traffic flow prediction and dynamic signal optimization method according to claim 1, characterized in that, The steps of learning the spatiotemporal matrix using a pre-defined spatiotemporal fusion model and outputting a prediction of traffic flow within a pre-defined time period using the learned spatiotemporal fusion model include: The spatial weights and periodic features in the spatiotemporal matrix are learned using a pre-defined spatiotemporal fusion model to obtain the learned spatiotemporal fusion model. The spatiotemporal fusion model outputs the predicted traffic flow for a future preset time period.

3. The AI-based urban traffic flow prediction and dynamic signal optimization method according to claim 1, characterized in that, The steps of adjusting the phase / green light duration corresponding to the traffic control node according to the global value score of the traffic control node and the time period characteristics include: Every first preset time interval, the traffic prediction results corresponding to the target area are received from the cloud, and the global value score of the traffic control node is updated according to the traffic prediction results. The control edge regenerates the signal control scheme for each traffic control node every second preset time interval. The signal control scheme includes the phase selection probability and the green light duration correction value.

4. An AI-based urban traffic flow prediction and dynamic signal optimization system, characterized in that, The system, applicable to the method of any one of claims 1-3, comprises: The data acquisition module is used to acquire real-time traffic data, historical flow data and external environment data of the target area to obtain multi-source data, and construct a spatiotemporal matrix corresponding to the target area based on the multi-source data; The prediction output module is used to learn the spatiotemporal matrix using a preset spatiotemporal fusion model, and output the prediction result of traffic flow within a preset time period through the learned spatiotemporal fusion model; wherein, the spatiotemporal fusion model is a joint architecture of graph convolutional network and Transformer network; The dynamic optimization module is used to dynamically adjust the traffic signal control parameters within the target area based on the prediction results using a multi-agent learning algorithm.

5. The AI-based urban traffic flow prediction and dynamic signal optimization system according to claim 4, characterized in that, The data acquisition module is used for: Real-time traffic data, historical flow data, and external environment data of the target area are acquired, and the real-time traffic data, historical flow data, and external environment data are fused to obtain multi-source data; Based on the multi-source data, the road network topology and time series data were extracted respectively; The intersections within the target area are modeled as graph nodes using the road network topology, spatial weights between adjacent intersections are extracted, and the time series data is periodically decomposed to extract periodic features from the time series data. Based on the spatial weights and the periodic features, a spatiotemporal matrix corresponding to the target region is constructed.

6. A readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the method described in any one of claims 1-3.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method according to any one of claims 1-3.

Images