CN119942815A

CN119942815A - A traffic artery coordinated control method and system

Info

Publication number: CN119942815A
Application number: CN202510003704.2A
Authority: CN
Inventors: 王志建; 张峻玮; 周清华; 刘小明; 陈智; 周慧娟; 郭伟伟; 吴文祥
Original assignee: North China University of Technology
Current assignee: North China University of Technology
Priority date: 2025-01-02
Filing date: 2025-01-02
Publication date: 2025-05-06

Abstract

The invention provides a traffic trunk coordination control method and a traffic trunk coordination control system, which belong to the field of road traffic control and comprise the steps of merging global information of traffic by introducing a graph neural network, coding a gating time sequence convolution network, obtaining global traffic prediction data according to coding states and merging and characteristics, mapping the global traffic prediction data into traffic state grades according to the prediction data, obtaining traffic flow and speed of local traffic, calculating fuzzy membership, calculating a weighting relation of traffic flows among nodes by utilizing a dynamic traffic flow weighting model, adjusting local green light timing, establishing a global-local combined objective function, inputting the traffic state grades and the local green light timing into the objective function, outputting global-local objective function values, selecting a control strategy with the smallest global-local objective function value, and outputting green light control duration. And a globally coordinated signal timing scheme is established, so that the traffic efficiency of the whole traffic trunk line is improved, and the traffic flow change and emergency can be responded quickly.

Description

Traffic trunk line coordination control method and system

Technical Field

The invention belongs to the field of road traffic control, and particularly relates to a traffic trunk line coordination control method and system.

Background

With the acceleration of the urban process and the continuous increase of the maintenance quantity of motor vehicles, the congestion problem of urban traffic trunks is increasing. The conventional traffic signal control method mainly includes fixed period control and inductive control. These methods play an important role in early urban traffic management, but with the acceleration of the urban process and the proliferation of traffic flow, the limitation thereof is increasingly prominent. Firstly, the adaptability is insufficient, and the fixed time period control is based on a preset signal timing scheme, so that the dynamic change of traffic flow cannot be responded in real time. Under the condition that traffic demand fluctuates greatly, the fixed timing can cause excessive saturation of certain road sections, and the resource waste of other road sections can not meet different demands of peak and off-peak periods. Secondly, it lacks global coordination, while inductive control is able to adjust signals according to local traffic conditions, it mainly lacks global optimization of the entire traffic network for individual intersections or local areas. This may lead to improvement of the local area at the expense of global efficiency, which occurs in the case of "this trade-off".

For the traffic conditions, expert scholars propose various traffic signal coordination control methods. The global coordination control method realizes the maximization of the overall traffic efficiency by carrying out macroscopic optimization on the whole traffic network. But this approach typically relies on accurate traffic models and large amounts of real-time data to predict traffic flow and adjust signal timing. However, the global method has high computational complexity, high implementation cost and high accuracy requirement on data. The local coordination control method focuses on the optimization of a single intersection or a local area, neglects the influence on the whole traffic network, and causes poor traffic efficiency in a global range.

Disclosure of Invention

In order to solve the problem of limitation of global coordination and local coordination in traffic control, the invention provides a traffic trunk coordination control method and system.

In order to achieve the above object, the present invention provides the following technical solutions:

a traffic trunk coordination control method, comprising the steps of:

The method comprises the steps of obtaining global information of a traffic trunk line in a target range, constructing a multi-source data analysis model based on a graph neural network and a gating time sequence convolution network, inputting the global information of traffic into the multi-source data analysis model, and outputting a global traffic flow predicted value, a density predicted value and a speed predicted value;

Obtaining traffic flow and vehicle speed of a local traffic trunk line in a target range, calculating fuzzy membership according to the traffic flow and the vehicle speed of the local traffic, calculating a weighted relation of traffic flows among nodes by using a dynamic traffic flow weighted model, and adjusting local green light timing by combining the weighted relation of the fuzzy membership and the traffic flows;

The traffic state grade and the local green light timing are input into the global-local combined objective function to output a plurality of global-local control parameters, a control strategy corresponding to the minimum global-local control parameter is selected, and the green light control duration is output according to the control strategy.

Preferably, the construction of the multi-source data analysis model based on the graph neural network and the gating time sequence convolution network inputs the global information of the traffic into the multi-source data analysis model, and outputs a global traffic flow predicted value, a density predicted value and a speed predicted value, specifically through the following steps:

Dividing the global information into node characteristics, edge characteristics and global structure characteristics;

And fusing the node characteristic fusion and the edge characteristic by adopting a graph neural network GNN to obtain a node fusion characteristic and an edge fusion characteristic, wherein the node fusion characteristic is calculated by the following formula:

A city trunk network G= (V, E), wherein V= { V ₁,v₂,...,v_N } represents N traffic nodes, E represents edges connecting the nodes, and the input feature of traffic node V _i is a vector composed of space-time data of S different data sources

Wherein q _i (t) represents the traffic flow at node v _i at time t, d _i (t) represents the vehicle density, v _i (t) represents the vehicle speed, and f _i (t) represents other data sources from outside including bus data and video monitoring data; A fused feature vector representing node v _i; phi (·) represents a nonlinear activation function, which is a ReLU function in the present invention; B _s represents bias vectors, F 'represents a uniform dimension of each data source after the transformation, and F=S×F' represents a feature dimension after fusion;

And when the global traffic network connectivity C is more than or equal to 0, carrying out edge feature fusion of the multi-source data, wherein the global traffic network connectivity is calculated by the following formula:

Wherein, the adjacency matrix a between nodes represents the topology structure of the road network, a _ij represents whether an edge exists between the nodes v _i and v _j, if there is a connection, a _ij =1, otherwise a _ij =0;

The edge fusion feature is calculated by the following formula:

Wherein e _uv∈R^D is represented as a fused feature vector of the edge (u, v), ψ (·) is represented as a nonlinear activation function; A feature transformation matrix representing the s' th edge data source; D ' is the unified dimension of each edge feature after transformation, D=S ' ×D ' is the dimension of the edge feature after fusion;

and performing time sequence coding on the node fusion characteristics by using a gating time sequence convolutional neural network GTCN, wherein the time sequence coding is performed by the following formula:

Wherein, The method comprises the steps of representing an initial hidden state of a node v, comprising time sequence information, wherein sigma (·) represents an activation function, the activation function is a ReLU, ζ (·) represents a gating function, the gating function is a sigmoid, K _τ,Q_τ∈R^K×F represents a convolution operation, K is a convolution kernel size, and b, c epsilon R ^F represents a bias vector; Represented as element level multiplication algorithm;

Calculating an edge characteristic weight value according to the time sequence coding state and the edge fusion characteristic, specifically by the following formula:

Where α _uv represents the attention weight of the edge (u, v); e _uv∈R^D represents the fusion feature of edges (u, v); Representing a transformation matrix; the method comprises the steps of representing an attention mechanism parameter vector, wherein gamma (·) represents LeakyReLU an activation function, I represents a vector splicing operation, N (v) represents a neighbor node set of a node v;

After computing the edge weights α _uv, messaging is performed and node states are updated, specifically by the following formula:

Wherein, W _h,W_s∈R^F×F represents a weight matrix of message passing and self-loop, sigma (·) is an activation function;

And integrating time sequence and space information to perform space-time attention fusion to obtain the fused representation of the node, wherein the fused representation is specifically represented by the following formula:

Wherein β _v represents the spatiotemporal attention coefficient of node v; Representing a spatiotemporal attention parameter vector; tan h (·) represents the hyperbolic tangent activation function; A fused representation representing node v;

the predicted traffic flow Q _v is output by the node, and the specific formula is as follows:

Wherein, The method comprises the steps of obtaining a predicted value K _v of the node V and a speed predicted value V _v, wherein the predicted value is represented by the node V, L represents the layer number of a network, f (·) represents an output mapping function, which refers to a linear layer in the invention.

Preferably, the mapping of the traffic flow predictor, the density predictor and the speed predictor to traffic status classes by fuzzy logic includes smoothness and congestion, calculated by the following formula:

S=f(Q_v,K_v,V_v);

Wherein S represents traffic state output, Q _v represents flow prediction value of the node V, K _v represents density prediction value of the node V, and V _v represents speed prediction value of the node V.

Preferably, calculating fuzzy membership according to the traffic flow and the speed of the local traffic, calculating the weighted relation of traffic flows between nodes by using a dynamic traffic flow weighted model, and adjusting local green light timing by combining the weighted relation of the fuzzy membership and the traffic flows, wherein the method specifically comprises the following steps:

calculating fuzzy membership according to the traffic flow and the speed of the local traffic, and passing through the following formula:

Wherein mu _Q,μ_V respectively represents fuzzy membership functions of the vehicle flow Q and the vehicle speed V, alpha and beta are respectively regulating parameters of the flow and the speed, and Q ⁰,V⁰ is a fuzzy threshold value;

calculating the weighted relation of traffic flow among nodes by using a dynamic traffic flow weighted model, wherein the dynamic traffic flow weighted model specifically comprises the following steps:

Wherein ω _uv represents the weighted traffic flow impact of node u on node v, Q _u represents the traffic flow at node u;

Q _uV_u represents the product of the traffic flow rate on the road section u, namely, under the timing of timing, the traffic flow of the node u and the vehicle speed jointly determine the traffic influence of the traffic flow of the node u on the downstream node V, and the control strategy of the signal lamp is dynamically adjusted according to the value and the traffic flow weighting in the neighbor node set;

And adjusting the partial green light timing by combining the weighted relation of the fuzzy membership and the traffic flow, and outputting a calculation formula of the partial green light timing G _g as follows:

G_g＝ω_uv(μ_Q+μ_V);

Wherein ω _uv represents the weighted traffic flow effect of node u on node V and μ _Q,μ_V represents the fuzzy membership functions of vehicle flow Q and vehicle speed V, respectively.

Preferably, the global-local joint objective function is established based on the global optimization objective function and the local optimization objective function, specifically:

The global optimization objective function J _global is:

Wherein n represents the number of trunk intersections, S _i represents the traffic state level 1 as clear and 0 as congestion, G _i represents the green light time of the ith intersection, queue _i represents the vehicle queuing length of the ith intersection, flow _i represents the traffic Flow of the ith intersection, omega ₁,ω₂ represents the global target weight, and the importance of the green light time and queuing efficiency is measured;

The local optimization objective function J _local is:

Wherein Delay _j represents the average Delay time of the vehicle at the jth intersection, S _j also represents the traffic state grade, 1 is smooth, 0 is congestion, G _j also represents the green light time, omega ₃,ω₄ represents the local target weight, and the importance of Delay time and green light allocation is measured;

The global-local joint optimization objective function is:

J_total＝δ·J_global+(1-δ)·J_local;

Wherein J _total represents a global and local combined optimization objective function, delta represents a global and local optimization balance coefficient, balance between global and local optimization objectives is adjusted, delta is more than or equal to 0 and less than or equal to 1;J _global, the global traffic optimization objective function reflects the efficiency of the whole traffic network, J _local represents a local signal control optimization objective function and reflects the specific signal control effect of each intersection.

Preferably, the method further comprises the step of setting the weight of the global-local traffic flow, specifically:

Wherein A _uv represents the traffic flow weight of the global node u and the local node V, Q _u represents the traffic flow of the global node u, V _u represents the vehicle speed of the global node u, Q _v represents the traffic flow of the local node V, V _v represents the vehicle speed of the local node V, and delta represents the balance coefficient of global and local optimization.

Preferably, the control strategy corresponding to the global-local control parameter with the smallest is selected, the green light control duration is output according to the control strategy, specifically, the control strategy corresponding to the smallest time in the global-local control parameters is selected, the green light time G _i of each intersection in the control strategy and the weight of the global-local traffic flow are calculated, and the green light control duration is output.

Preferably, the weights of the green light time G _i and the global-local traffic flow of each intersection in the control strategy are calculated, specifically by the following formula:

Wherein G' _g represents the green light duration of the local node v, G _g represents the initial green light duration of the local node v, Δg represents the adjustment increment of the green light duration, and Δg=g _g-G_g;Σ_u∈N(v)A_uv represents the accumulation of global-local traffic flow weights, reflecting the influence of the peripheral traffic flow on the current road section.

The invention also provides a traffic trunk line coordination control system, which specifically comprises:

the system comprises a global data acquisition and analysis module, a multi-source data analysis model, a fuzzy logic mapping module and a traffic state grade mapping module, wherein the global data acquisition and analysis module is used for acquiring global information of a traffic trunk line in a target range, the global information specifically comprises vehicle data and road condition data, the multi-source data analysis model is constructed based on a graphic neural network and a gating time sequence convolution network, the global information of traffic is input into the multi-source data analysis model, global traffic flow predicted values, density predicted values and speed predicted values are output, and the traffic flow predicted values, the density predicted values and the speed predicted values are mapped into traffic state grades through the fuzzy logic.

The local signal control module is used for acquiring the traffic flow and the vehicle speed of a local traffic trunk line in a target range, calculating fuzzy membership according to the traffic flow and the vehicle speed of the local traffic, calculating the weighted relation of traffic flows among nodes by using a dynamic traffic flow weighted model, and adjusting local green light timing by combining the weighted relation of the fuzzy membership and the traffic flows.

The global-local coordination scheduling module is used for establishing a global-local combined objective function based on the global optimization objective function and the local optimization objective function, inputting the global-local combined objective function when the traffic state level and the local green light are time-matched, outputting a plurality of global-local control parameters, selecting a control strategy corresponding to the minimum global-local control parameters, and outputting the green light control duration according to the control strategy.

The traffic trunk line coordination control method provided by the invention has the following beneficial effects:

The method is characterized in that a multisource data analysis model is built based on a graph neural network and a gating time sequence convolution network, global information of an acquired traffic trunk is input into the multisource data analysis model to obtain a global traffic flow predicted value, a density predicted value and a speed predicted value, and fusion of characteristics in a unified space is achieved through characteristic transformation and splicing of multisource data, so that complex space-time processing of the model is improved. And obtaining the traffic flow and the speed of the local traffic, calculating the membership degree, calculating the weighted relation of traffic flow among nodes by using a dynamic traffic flow weighted model, and adjusting the local green light timing by combining the fuzzy membership degree and the weighted relation of the traffic flow. And outputting an optimal green light duration control strategy based on the global-local combined objective function according to the predicted value of the global traffic flow and the local control strategy. And the signal lamp timing scheme is adaptively adjusted, so that the dynamic optimization of global-local signal control is realized, and the waiting time and traffic jam of the vehicle are effectively reduced.

Drawings

In order to more clearly illustrate the embodiments of the present invention and the design thereof, the drawings required for the embodiments will be briefly described below. The drawings in the following description are only some of the embodiments of the present invention and other drawings may be made by those skilled in the art without the exercise of inventive faculty.

Fig. 1 is a flow chart of a traffic trunk coordination control method of the present invention.

Fig. 2 is a global traffic network diagram of an embodiment of the present invention.

Fig. 3 is a diagram of a trunk coordination traffic network in accordance with an embodiment of the present invention.

Fig. 4 is a diagram showing the control effect of the local signal lamp according to the embodiment of the present invention.

Fig. 5 is a diagram of a global local coordination effect according to an embodiment of the present invention.

Fig. 6 is a graph of the global and local traffic flow optimization before and after the change in the embodiment of the invention.

Fig. 7 is a diagram showing latency variations before and after signal optimization in an embodiment of the present invention.

Fig. 8 is a diagram of the signal lamp timing optimization before and after the signal lamp timing optimization in the embodiment of the invention.

FIG. 9 is a graph of traffic flow thermodynamic diagram after global local coordination in an embodiment of the present invention.

Fig. 10 is a thermodynamic diagram of traffic flow in an embodiment of the invention.

Detailed Description

In order that those skilled in the art will better understand the technical solutions of the present invention and implement it, the present invention will be described in detail with reference to the drawings and the specific embodiments. The following examples are only for more clearly illustrating the technical aspects of the present invention, and are not intended to limit the scope of the present invention.

Examples

The invention provides a traffic trunk line coordination control method, which is shown in fig. 1 and specifically comprises the following steps:

S1, acquiring global information of traffic. Global information of the trunk traffic is collected through various data sources, and mainly comprises floating car data, drive test sensor data, video monitoring data and public traffic data. The vehicle floating data are obtained through vehicle satellite positioning equipment, the position information, the speed and the acceleration of the vehicle can be reflected in real time, the vehicle floating data are important sources of dynamic traffic flow data, road side sensor data are the quantity, the speed and the lane occupation conditions of passing vehicles detected by sensors arranged along a trunk road, video monitoring data are traffic state information in video streams are extracted through cameras arranged at main intersections, the quantity, the type and the running direction of the vehicles are identified by utilizing a computer vision technology, and public traffic data are operation data from vehicles such as buses, taxis and shared buses, so that the state of urban traffic flow can be comprehensively estimated.

S2, constructing a multi-source data analysis model based on the graph neural network and the gating time sequence convolution network, inputting global traffic information into the multi-source data analysis model, outputting a global traffic flow predicted value, a global density predicted value and a global speed predicted value, and mapping the traffic flow predicted value, the global density predicted value and the global speed predicted value into traffic state grades through fuzzy logic. The global traffic network diagram is shown in fig. 2.

The global traffic optimization objective function specifically comprises:

Wherein global represents a global optimization objective function, T _i represents a vehicle passing time of an ith road segment, W _i represents an average vehicle waiting time of the ith road segment, C _i represents a vehicle flow passing rate of the ith road segment, and lambda ₁,λ₂,λ₃ represents a target weight for measuring priorities among different targets. The traffic flow prediction thermodynamic diagram is shown in fig. 10.

S21, different data sources have heterogeneity and inconsistency, and a multisource data fusion algorithm is needed to process the heterogeneous data sources. The method comprises the steps of fusing multisource space-time data by adopting a Graph Neural Network (GNN), establishing a multisource data analysis model, processing a topological structure in a traffic network, integrating information from different sensors and data sources, and considering multi-element characteristics of nodes and edges, space-time sequence information and a space topological structure. The relationship of nodes and traffic of the trunk coordinated traffic network is shown in fig. 3.

For a city trunk network g= (V, E), where v= { V ₁,v₂,...,v_N } represents N traffic nodes (intersections in the trunk or segments between adjacent intersections), E represents edges connecting these nodes (specifically represented as interconnections between the connecting segments), the input features of traffic node V _i are vectors composed of spatio-temporal data of S different data sourcesThe multisource node fusion formula is as follows:

Where q _i (t) represents the traffic flow at node vi at time t, d _i (t) represents the vehicle density, v _i (t) represents the vehicle speed, and f _i (t) represents other data sources from outside (e.g., bus data, video monitoring data, etc.); A fused feature vector representing node v _i; phi (·) represents a nonlinear activation function, which is a ReLU function in the present invention; The method comprises the steps of representing feature transformation matrixes of S data sources, b _s representing bias vectors, F 'representing the unified dimension of each data source after the feature is transformed, and F=S×F' representing the feature dimension after fusion.

The adjacency matrix a between nodes represents the topology of the road network, a _ij represents whether an edge exists between nodes v _i and v _j, a _ij =1 if there is a connection, and a _ij =0 otherwise. The influence degree among different road segments is measured by combining with a complex network theory, global decision of the whole traffic flow control is optimized, and the pre-judgment is made, wherein the formula is as follows:

Wherein, the adjacency matrix a between nodes represents the topology structure of the road network, a _ij represents whether an edge exists between the nodes v _i and v _j, if there is a connection, a _ij =1, otherwise a _ij =0.

S22, performing node fusion, and if the network connectivity C is more than or equal to 0, performing edge feature fusion of the multi-source data. The image neural network fusion multi-source space-time data is subjected to multi-source node feature fusion, multi-source edge feature fusion is also required, and for the edge (u, v) E, edge features from S' data sources are providedWhere S '=1, 2,..s', then the edge feature fusion formula is as follows:

Wherein e _uv∈R^D is represented as a fused feature vector of the edge (u, v), ψ (·) is represented as a nonlinear activation function; a feature transformation matrix representing the s' th edge data source; Expressed as a bias vector, D ' expressed as a transformed unified dimension for each edge feature, and d=s ' ×d ' expressed as a fused edge feature dimension.

Data of the nearest Γ time steps are encoded taking into account the spatio-temporal dynamics of the node characteristics, while defining a time window Γ= { T-t+1..the, T }, the characteristics of the node v being, for each time step τ e ΓThe gating sequential convolutional neural network GTCN is adopted for sequential encoding, and the formula is as follows:

Wherein, The method comprises the steps of representing an initial hidden state of a node v and comprising time sequence information, wherein sigma (-) represents an activation function, zeta (-) represents a gating function, the gating function is sigmoid, K _τ,Q_τ∈R^K×F represents a convolution operation, K is a convolution kernel size, and b, c epsilon R ^F represents a bias vector; represented as element level multiplication algorithms.

S23, calculating the relevance among nodes of the weight alpha _uv of the multi-element feature calculation edge (u, v), wherein the relevance is specifically calculated by the following formula:

Where α _uv represents the attention weight of the edge (u, v); e _uv∈R^D represents the fusion feature of edges (u, v); Representing a transformation matrix; represents an attention mechanism parameter vector, gamma (·) represents LeakyReLU an activation function, i represents a vector concatenation operation, and N (v) represents a set of neighbor nodes of node v.

And carrying out message transmission according to the weight and updating the node state, specifically by the following formula:

Wherein, Representing hidden states of the layer 1 node v, W _h,W_s∈R^F×F representing weight matrices for message passing and self-loop, σ (·) being an activation function.

S24, integrating time sequence and space information to perform space-time attention fusion to obtain the fused representation of the nodes, wherein the fused representation is specifically represented by the following formula:

Wherein β _v represents the spatiotemporal attention coefficient of node v; Representing a spatiotemporal attention parameter vector; tan h (·) represents the hyperbolic tangent activation function; Representing a fused representation of node v.

Wherein, Representing the predicted value of the node v, L representing the number of layers of the network, f (·) representing the output mapping function, referred to herein as the linear layer.

The loss function is defined in the multi-element space-time data fusion model to train the model, and an Adam optimizer is adopted to update model parameters, and the specific formula is as follows:

Q _v represents the true value of the node v, lambda regularization coefficient, all trainable parameter sets of the theta model, eta learning rate; Representing the gradient of the loss function versus the parameter.

And S25, by combining the traditional data (road side sensor) with the dynamic data (vehicle track), the real-time performance and accuracy of the traffic flow state are improved, the flow predicted value Q _v of the node V is finally obtained, and the density predicted value K _v and the speed predicted value V _v of the node V are obtained in the same way. Then, traffic state evaluation and output are carried out, and according to the analyzed data, a fuzzy logic system is adopted for traffic state evaluation, and two fuzzy rules are defined: traffic status is "congested" if the predicted flow value is large, and "unblocked" if the predicted flow value is small. And carrying out traffic state judgment by combining the two rules with a fuzzy membership function, wherein the formula is as follows:

S=f(Q_v,K_v,V_v);

Through fuzzy logic, the system maps successive traffic parameters to different traffic state levels ("clear" or "congested") providing basis for global and local control.

S3, obtaining the traffic flow and the vehicle speed of the local traffic, calculating the fuzzy membership according to the traffic flow and the vehicle speed of the local traffic, calculating the weighted relation of traffic flows among nodes by using a dynamic traffic flow weighted model, and adjusting the local green light timing by combining the weighted relation of the fuzzy membership and the traffic flows.

And adjusting a signal lamp timing strategy according to the traffic state, and controlling the local signal. The local signal control optimization objective function specifically comprises the following steps:

Wherein J _local denotes a local optimization objective function, W _t denotes an average waiting time of the vehicle at time t, P _t denotes an average number of stops of the vehicle at time t, C _t denotes a vehicle passing rate at time t, and lambda ₁,λ₂,λ₃ denotes a weight parameter.

S31, calculating fuzzy membership according to traffic flow and speed of local traffic, specifically by the following formula:

wherein mu _Q,μ_V respectively represents fuzzy membership functions of the traffic flow Q and the vehicle speed V, alpha and beta are respectively adjustment parameters of the traffic flow and the vehicle speed, Q ⁰,V⁰ is a fuzzy threshold value, and a demarcation point of a normal state is defined for distinguishing low, medium and high traffic flows or the vehicle speed, and the adjustment parameters alpha and beta control the steepness degree of the fuzzy membership functions, namely the sensitivity of the change of input variables (the traffic flow and the vehicle speed) to the membership degree. In the invention, the value alpha=0.5 and the value beta=0.03 are determined by trial and error due to global-local coordination control.

S32, calculating the weighted relation of traffic flows among nodes by using a dynamic traffic flow weighted model, wherein the dynamic traffic flow weighted model specifically comprises the following steps:

Wherein ω _uv represents the weighted traffic flow effect of node u on node V, Q _u represents the traffic flow at node u, V _u represents the average speed at node u, Q _uV_u represents the traffic flow product on road section u, i.e. under the timing of timing, the traffic flow and speed of node u together determine the traffic effect on downstream node V, the traffic flow weighting in the set of neighboring nodes is based on this value, the control strategy of the signal lamp is dynamically adjusted, N (V) represents the set of neighboring nodes of node V, i.e. all road sections or intersections connected to node V, and Σ _k∈N(v)Q_kV_k represents the total traffic flow in the neighboring nodes of node V.

S33, adjusting the partial green light timing by combining the weighted relation of the fuzzy membership and the traffic flow, and outputting a green light duration G _g according to the calculation formula:

G_g＝ω_uv(μ_Q+μ_V);

Wherein ω _uv represents the weighted traffic flow effect of node u on node V and μ _Q,μ_V represents the fuzzy membership functions of vehicle flow Q and vehicle speed V, respectively. The local signal control effect is shown in fig. 4.

S4, coordinating global and local signal timing to ensure smooth overall traffic flow of the urban trunk line. The global-local combined objective function is established according to the global optimization objective function and the local optimization objective function, and specifically comprises the following steps:

J_total＝δ·J_global+(1-δ)·J_local;

Wherein J _total represents a global and local combined optimization objective function, delta represents a global and local optimization balance coefficient, delta is more than or equal to 0 and less than or equal to 1;J _global, the global traffic optimization objective function reflects the efficiency of the whole traffic network, and J _local represents a local signal control optimization objective function and reflects the specific signal control effect of each intersection.

S41, classifying J _total into a plurality of grades, and evaluating the effect of the optimization result. The following evaluation criteria were defined:

J _total is more than or equal to 0 and less than or equal to 500, and the traffic flow is almost not delayed, the signal timing scheme at the intersection is reasonable, the traffic flow is close to the theoretical maximum value, and the traffic flow is almost not jammed.

J _total >500, which indicates that traffic flow is seriously jammed, the signal timing of the intersection fails to effectively respond to traffic flow change, and the system is not smooth to operate and jam is generated.

The data are expressed in the form of a matrix, and are used as input matrixes of the joint optimization model, and the input matrixes are generalized as follows:

Assuming that the trunk line of a city has 4 intersections (numbers 1 to 4), the input data of each intersection includes a traffic state level, which is mapped from traffic parameters (traffic flow, vehicle speed) through fuzzy logic, namely, "unblocked" is 1, and "blocked" is 0. Green time-green time duration (units: seconds) for each intersection provided according to the local traffic timing.

Table 1 global-local example dataset table

The data set structure states that the traffic status levels, e.g., the status of intersections 1 and 3, are "clear" (1) and the status of intersections 2 and 4 are "congested" (0). The green time is a specific signal lamp duration, for example, the green time of the intersection 2 is 45 seconds, and the intersection 4 is 50 seconds.

The data are represented in the form of a matrix:

Wherein each row corresponds to the input data of an intersection, and the columns represent different input features, the first column is the traffic state level S _i and the second column is the green time G _j.

The joint objective function is expanded as follows:

the joint objective function needs to meet the following constraints:

Green light time range constraint: traffic state adjustment consistency constraints: epsilon is a very small value, queuing traffic constraint:

The same main traffic needing to be coordinated is input into the joint objective function J _total by different data matrixes X in the data source. In the same coordination trunk line, different control schemes have different traffic states and green light time, data matrixes corresponding to the different control schemes input J _total calculated by the combined objective function correspond to different values, and according to the rule provided by the invention, which scheme calculates the smallest value of J _total, which scheme is the best. And (3) taking out the second column in the data matrix X of the best scheme, namely, the green light time G _i obtained by solving the objective function of the main line for coordinating each intersection.

S42, constructing global and local traffic flow coordination and calculating based on the traffic flow weights of the nodes and the edges. A global-local traffic flow weight matrix a _uv is defined to reflect the traffic flow relationship between the global node u and the local node v, specifically as follows:

Wherein A _uv represents the traffic flow weight of the global node u and the local node V, Q _u represents the traffic flow of the global node u, V _u represents the vehicle speed of the global node u, Q _v represents the traffic flow of the local node V, V _v represents the vehicle speed of the local node V, gamma represents the coordination coefficient of the global and local traffic flows, and N (V) represents the local node V and the set of neighboring nodes.

S43, the system can dynamically adjust the signal lamp allocation of each intersection through the global and local weights obtained through calculation, the final green light duration adjustment formula of each intersection is shown as follows, G _i in the case of i=1, 2 is brought in when the local green light is timed, and single final adjustment is carried out according to the edges and the nodes, the green light time of the single intersection on the trunk line is finally adjusted, namely the system returns to the local coordination control, different X inputs global-local joint objective functions to enable the calculation value of the objective function to be minimum, and the best scheme is obtained. And finally, performing single-port final tuning according to the green light time G _i of each single-port in the matrix X. The global and local coordination effect is shown in fig. 5, and the green light duration is specifically adjusted by the following formula:

Wherein G' _g represents the green light time of the local node v, G _g represents the initial green light time of the local node v, ΔG represents the adjustment increment of the green light time, and Σ _u∈N(v)A_uv represents the accumulation of global-local traffic flow weights, reflecting the influence of the peripheral traffic flow on the current road section.

As shown in fig. 6, 7, 8 and 9, the present invention is used for the comparison of the traffic trunk coordination control optimization before and after.

In order to solve the defects of the existing traffic signal control method in the aspects of global coordination, real-time adaptability and multi-objective optimization, the high-efficiency and intelligent management of the traffic trunk is realized. A multi-level control method integrating global optimization and local dynamic adjustment is constructed. And the global coordinated signal timing scheme is established by carrying out optimized scheduling of the global traffic network on a macroscopic level so as to improve the traffic efficiency of the whole traffic trunk. Meanwhile, on the microscopic level, real-time traffic data is utilized to dynamically adjust local intersections and road sections, so that the change of traffic flow and emergency are responded quickly, and good operation of local traffic conditions is ensured. The method has the following beneficial effects:

(1) Based on GTCN and multi-source spatio-temporal data joint analysis of the graph neural network, a combination of a gating time sequence convolution network (GTCN) and the Graph Neural Network (GNN) is innovatively introduced in a global data analysis and acquisition module. Through GTCN, the system can efficiently extract time sequence dependent characteristics of traffic data in the time sequence and capture dynamic changes of global traffic flow. Meanwhile, the GNN model is used for processing the complex network topological relation of the space dimension and analyzing the traffic flow association between different road sections and intersections. The two are combined, so that the system not only can accurately predict the space-time evolution of traffic flow, but also can process a complex global traffic network topology structure, and the prediction accuracy and the real-time performance of traffic data are greatly improved.

(2) Based on the reinforcement learning and the local signal dynamic optimization control of the TCN, the reinforcement learning is combined with a time sequence convolution network (TCN), so that a brand new local traffic signal optimization strategy is formed. The reinforcement learning helps the system to adjust according to traffic flow states in a real-time environment by adaptively adjusting timing of the signal lamps. The TCN predicts and analyzes the historical data of the local traffic flow through the strong time sequence modeling capability, so that the accuracy and the response speed of signal lamp control are further improved. The combination realizes the dynamic optimization of local signal control, and can effectively reduce the waiting time of the vehicle and traffic jam.

(3) The invention provides a global-local combined optimization model, which balances the macro control of global traffic flow and the fine regulation of local signals by jointly optimizing global and local traffic flow. The model integrates global GTCN and local TCN, constructs a space-time combined multi-objective optimization function, and carries out comprehensive optimization solving on indexes such as global traffic flow, local signal lamp timing, waiting time and the like. The traffic light timing scheme can be adjusted in a self-adaptive mode, and global and local coordination scheduling of traffic flows is achieved.

The invention improves the overall efficiency and stability of the traffic flow, and can improve the overall efficiency of the traffic flow and reduce the bottleneck and congestion in the traffic network by combining the global and local optimization strategies. The optimized traffic signal can be dynamically adjusted according to the real-time flow, so that the stability of the traffic network is improved, and traffic delay in peak time is reduced. The invention realizes personalized adjustment at each intersection through local signal optimization, thereby reducing traffic delay and queuing conditions, especially at high-flow intersections. The local optimization ensures the maximization of the traffic capacity of the intersection by self-adaptive signal adjustment, and improves the local traffic fluency. The invention can effectively avoid the phenomenon of 'inter-traffic elimination', ensure the smooth global traffic and fully guarantee the traffic capacity of a local intersection through the coordination of global and local signal optimization. The optimization of traffic flow is not limited to a certain local area only, but an efficient operation of the entire traffic network is achieved.

The global data acquisition and analysis module is used for acquiring global information of a traffic trunk line in a target range, wherein the global information specifically comprises vehicle data and road condition data, a multi-source data analysis model is built based on a graph neural network and a gating time sequence convolution network, the global information of traffic is input into the multi-source data analysis model, global traffic flow predicted values, density predicted values and speed predicted values are output, and the traffic flow predicted values, the density predicted values and the speed predicted values are mapped into traffic state grades through fuzzy logic.

The local signal control module is used for acquiring the traffic flow and the vehicle speed of the local traffic trunk line in the target range, calculating the fuzzy membership according to the traffic flow and the vehicle speed of the local traffic, calculating the weighted relation of traffic flows among nodes by utilizing the dynamic traffic flow weighted model, and adjusting the local green light timing by combining the weighted relation of the fuzzy membership and the traffic flows.

The global-local coordination scheduling module is used for establishing a global-local combined objective function based on the global optimization objective function and the local optimization objective function, inputting the traffic state grade and the local green light timing into the global-local combined objective function, outputting a plurality of global-local control parameters, selecting a control strategy corresponding to the minimum global-local control parameters, and outputting the green light control duration according to the control strategy.

The system integrates multi-source traffic data by the global data acquisition and analysis module to construct a global traffic state sensing system, so that the dynamic monitoring and prediction capability of the whole traffic network is improved. Through data fusion and complex network modeling, the change trend of the global traffic flow can be perceived in real time, traffic jam is reduced, traffic efficiency is improved, meanwhile, uncertainty caused by traffic fluctuation is reduced through optimizing global traffic stability, and the efficiency and stability of the global traffic flow are improved.

And in the local signal control module, the timing strategy of the signal lamp is dynamically adjusted at each intersection or road section according to the real-time traffic state. The self-adaptive algorithm combining reinforcement learning and fuzzy control can automatically optimize the green light duration of the signal lamp according to key indexes such as traffic flow, vehicle waiting time and the like, reduce the waiting time and parking times of the vehicle and improve the traffic efficiency of local traffic.

And the global-local coordination scheduling module organically combines the global traffic control targets with the local traffic control targets. The balance between the global traffic fluidity and the local signal control can be found, the overall efficiency maximization of the global traffic network can be ensured, the personalized requirement of the local signal control can be met, and the efficient traffic flow regulation and control can be realized. And the method can realize global and local real-time coordination and emergency dispatch and can cope with emergencies and traffic anomalies. By means of real-time data feedback and dynamic optimization adjustment, the system can rapidly detect abnormal conditions, emergency signal scheduling is conducted on the overall and local levels, rapid recovery and distribution of traffic flows are guaranteed, emergency conditions such as traffic accidents and road construction are effectively treated, and self-adaptive emergency management is achieved.

The modules in the traffic trunk coordination control system can be fully or partially realized by software, hardware and combinations thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.

It will be apparent to those skilled in the art that embodiments of the present invention may provide a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flowchart and/or block of the flowchart illustrations and/or block diagrams, and combinations of flowcharts and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

It should be noted that the above-described embodiments will enable those skilled in the art to more fully understand the invention, but do not limit it in any way. Therefore, although the present invention has been described in detail with reference to the present specification and examples, it will be understood by those skilled in the art that the present invention may be modified or substituted for others, and all such modifications and improvements which do not depart from the spirit and scope of the present invention are intended to be included in the scope of the present invention. Any reference sign in a claim should not be construed as limiting the claim concerned. Any simple modification or equivalent substitution of the technical solutions that can be obviously obtained by those skilled in the art within the technical scope of the present disclosure fall within the protection scope of the present invention.

Claims

1. The traffic trunk coordination control method is characterized by comprising the following steps of:

2. The traffic trunk coordination control method according to claim 1, wherein the constructing a multi-source data analysis model based on a graph neural network and a gating time sequence convolution network inputs global information of the traffic into the multi-source data analysis model, and outputs a global traffic flow predicted value, a density predicted value and a speed predicted value, specifically by the following steps:

The edge fusion feature is calculated by the following formula:

Wherein e _uv∈R^D is represented as a fused feature vector of the edge (u, v), ψ (·) is represented as a nonlinear activation function; The feature transformation matrix of the S 'th side data source is represented, D _s'∈R^D′ is represented as a bias vector, D' is represented as a unified dimension after each side feature is transformed, and D=S '×D' is represented as a fused side feature dimension;

Where α _uv represents the attention weight of the edge (u, v); Representing hidden states of the nodes u and v of the first layer, e _uv∈R^D representing fusion features of edges (u, v), W ₁,W₂∈R^F'×F,W₃∈R^F'×D representing a transformation matrix, a epsilon R ^3F' representing an attention mechanism parameter vector, gamma (·) representing LeakyReLU an activation function, ||representing a vector concatenation operation, and N (v) representing a set of neighbor nodes of the node v;

Wherein, beta _v represents the space-time attention coefficient of the node v, u epsilon R ^F' represents the space-time attention parameter vector; tan h (·) represents the hyperbolic tangent activation function; A fused representation representing node v;

3. The traffic trunk coordination control method according to claim 2, wherein the traffic flow predicted value, the density predicted value, and the speed predicted value are mapped to traffic state levels by fuzzy logic, the traffic state levels including clear and congestion, calculated by the following formula:

S=f(Q_v,K_v,V_v);

4. The traffic trunk coordination control method according to claim 3, wherein the fuzzy membership is calculated according to the traffic flow and the vehicle speed of the local traffic, the weighted relation of traffic flows between nodes is calculated by using a dynamic traffic flow weighted model, and the local green light timing is adjusted by combining the weighted relation of the fuzzy membership and the traffic flows, specifically comprising the following steps:

G_g＝ω_uv(μ_Q+μ_V);

5. The traffic trunk coordination control method according to claim 4, wherein the global-local joint objective function is established based on a global optimization objective function and a local optimization objective function, specifically:

The global optimization objective function J _global is:

The local optimization objective function J _local is:

The global-local joint optimization objective function is:

J_total＝δ·J_global+(1-δ)·J_local;

6. The traffic trunk coordination control method according to claim 5, further comprising setting weights of global-local traffic flows, specifically:

7. The method for coordinated control of a traffic trunk according to claim 6, wherein the control strategy corresponding to the smallest global-local control parameter is selected, a green light control duration is output according to the control strategy, specifically, the control strategy corresponding to the smallest time among the plurality of global-local control parameters is selected, the green light time G _i of each intersection in the control strategy and the weight of the global-local traffic flow are calculated, and the green light control duration is output.

8. The method according to claim 6, wherein the weights of the green time G _i and the global-local traffic flow of each intersection in the control policy are calculated by the following formula:

Wherein G' _g represents the green light duration of the local node v, G _g represents the initial green light duration of the local node v, Δg represents the adjustment increment of the green light duration, and Δg=g _g-G_g;∑_u∈N(v)A_uv represents the accumulation of global-local traffic flow weights, reflecting the influence of the peripheral traffic flow on the current road section.

9. A traffic trunk coordination control system, comprising:

The system comprises a global data acquisition and analysis module, a multi-source data analysis model, a fuzzy logic mapping module, a traffic state grade mapping module and a traffic state grade mapping module, wherein the global data acquisition and analysis module is used for acquiring global information of a traffic trunk line in a target range, the global information specifically comprises vehicle data and road condition data, the global information of traffic is input into the multi-source data analysis model based on a graphic neural network and a gating time sequence convolution network, and global traffic flow prediction value, density prediction value and speed prediction value are output;

The local signal control module is used for acquiring the traffic flow and the vehicle speed of a local traffic trunk line in a target range, calculating fuzzy membership according to the traffic flow and the vehicle speed of the local traffic, calculating the weighted relation of traffic flows among nodes by using a dynamic traffic flow weighted model, and adjusting local green light timing by combining the weighted relation of the fuzzy membership and the traffic flows;