Smart city traffic planning method and system based on deep learning

Abstract

The invention discloses a smart city traffic planning method and system based on deep learning, and relates to the technical field of city traffic management, and the method comprises the steps: recognizing an activity region of a traffic disturbance activity, constructing a disturbance region graph Ga, collecting vehicle driving related data and lamp control cycle data, and constructing a standardized data set H; coupling modeling is carried out on path length change and direction deviation of the traffic driving path before and after the traffic disturbance activity so as to quantify the traffic state of the activity area, a path disturbance deviation coefficient Py is acquired, the traffic state of the activity area is evaluated in combination with a path disturbance threshold Pyyz, and when the traffic state is in an abnormal state, the traffic disturbance deviation coefficient Py is calculated. According to the method, signal lamp cycle rhythm analysis is carried out, a phase dislocation propagation index Xw is obtained, and coupling modeling is carried out to obtain a comprehensive rhythm adaptation index Zh, so that the risk level of a traffic signal control system is divided, a system adjustment instruction is generated, intelligent adjustment is carried out, and the intelligent, precise and real-time level of smart city traffic planning is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of urban traffic management, in particular to a smart city traffic planning method and system based on deep learning. BACKGROUND

[0002] In the city area where temporary activities such as holiday peak, concert, sports event, and large-scale exhibition are frequently held, there are often phenomena such as short-term dramatic fluctuation of regional traffic capacity, significant increase of path detour, and mismatch of signal response, especially in the dense intersection and asymmetric traffic network structure, the existing traffic control system is prone to problems such as imbalance of traffic link scheduling, misplacement of green light resource release, and polarization of queue length. The current technology mainly relies on manual prediction of activity area and road section restriction or manual adjustment of cycle, but it is difficult to cover all the propagation paths, and it is difficult to identify and time-domain correct the mismatch of linked signals, resulting in lag of traffic capacity recovery time and secondary diffusion of traffic disturbance.

[0003] The existing urban traffic control strategy is mainly based on static regional configuration and historical average flow distribution, and lacks the ability to identify the propagation path of signal rhythm mismatch caused by temporary disturbance. First, the rhythm control of traffic signal is mainly single-node cycle adjustment or stage response scheme, which is difficult to model based on the path deflection trend and queue propagation trend of vehicles in the disturbance area. Second, the existing control system generally uses fixed signal plan, and the cycle adjustment trigger condition is lagging, only when the vehicle delay is significantly over-standard, the queue processing flow is triggered, resulting in insufficient system adjustment ability and response granularity, and it is difficult to form a system repair mechanism for signal rhythm imbalance in the disturbance area.

[0004] The above problems have existed for a long time, resulting in that in the traffic disturbance scene, although the local area has carried out flow guidance or traffic control, the accumulation of signal rhythm mismatch is difficult to repair in time, thus forming a mismatch diffusion main chain, inducing large-scale signal rhythm imbalance chain response, and finally forming phenomena such as periodic instability of traffic, delay of intersection green light release, and dynamic decline of traffic capacity. In addition, the existing technology lacks a signal cycle self-learning adjustment mechanism based on traffic path disturbance behavior feedback, so that the system is difficult to accumulate rhythm response experience in multiple activity disturbances, and it is also difficult to adaptively generate adjustment suggestions combined with historical rhythm mismatch data, and it is difficult to realize the optimization evolution of rhythm reconstruction. Therefore, there is an urgent need for a smart city traffic planning method and system triggered by disturbance events. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a smart city traffic planning method and system based on deep learning, which solves the problems mentioned in the background art.

[0006] To achieve the above object, the present application is implemented by the following technical solutions: comprising the following steps:

[0007] S1, determining the activity area of the traffic disturbance activity through the urban social media platform, to construct the disturbance area map Ga, and collecting the vehicle driving related data and the light control cycle data in the disturbance area map Ga, to construct the standardized data set H;

[0008] S2, according to the disturbance area map Ga, identifying the traffic driving path, and analyzing the path length change and direction deviation of the vehicle before and after the traffic disturbance activity, to evaluate the traffic state of the activity area, and when the traffic state of the activity area is in an abnormal state, analyzing the signal light cycle rhythm;

[0009] S3, when the traffic state of the activity area is in an abnormal state, calculating the rhythm center point difference of the traffic signal light according to the light control cycle data, and quantifying the rhythm consistency and response coordination between the traffic signal lights;

[0010] S4, according to the disturbance area map Ga, using the K shortest path algorithm to identify the total number M of traffic driving paths, and analyzing the running state of the traffic signal control system in the activity area, to divide the risk level and generate system adjustment instructions of different risk levels;

[0011] S5, according to the system adjustment instruction, intelligently adjusting the traffic signal control system.

[0012] Preferably, the specific steps of S1 include:

[0013] S11, according to the crawler grabbing tool Scrapy, grabbing the traffic disturbance activity related fields on the urban social media platform, including text content, publishing time and location information, to identify the traffic disturbance activity in the city, and combining the city map database, extracting the boundary point coordinates of the traffic disturbance activity, and according to the boundary point coordinates of the traffic disturbance activity, using the convex hull algorithm, sorting all the boundary point coordinates by polar angle, starting from the leftmost boundary point, connecting in the counterclockwise direction in turn, forming a closed figure area, that is, the activity area of the traffic disturbance activity, wherein the boundary point coordinates refer to the GPS geographical position point coordinates used to describe the boundary of the traffic disturbance activity affected area, which are obtained by analyzing the location information of the traffic disturbance activity;

[0014] S12, setting each traffic signal lamp as a traffic signal lamp node in the activity area of the traffic disturbance activity, traversing the entire activity area of the traffic disturbance activity, constructing a traffic signal lamp node set, calling a signal control platform log to obtain lamp control period data of the traffic signal lamp node, wherein the lamp control period data includes a green light start time Tstart and a green light end time Tend, and obtaining GPS coordinate data of each traffic signal lamp node according to a city map database to construct a disturbance area graph Ga, wherein the disturbance area graph Ga=(Na, Ea), Na represents the traffic signal lamp node set, and Ea represents a road connection edge set;

[0015] S13, setting a plurality of data collection intervals in the activity area, and deploying a radar speed measurement device and a high-definition portal camera in each data collection interval to obtain vehicle driving related data in the activity area in real time, wherein the vehicle driving related data includes a vehicle driving trajectory and a vehicle driving distance d;

[0016] The vehicle driving trajectory refers to the driving route of the vehicle on the traffic driving road, including the trajectory point coordinates of the vehicle on the traffic driving road, which is obtained by identifying and obtaining the high-definition portal camera;

[0017] The vehicle driving distance d refers to the actual trajectory length of the vehicle moving along the traffic driving route in the activity area, and the through time and speed of the vehicle at each coordinate point are collected by the radar speed measurement device, and the speed integral calculation is obtained, wherein the radar speed measurement device is installed on the roadside signal lamp pole, monitoring pole and power pole by using a fixed support.

[0018] Preferably, the S1 specific steps further comprise:

[0019] S14, data preprocessing is performed on the collected lamp control period data and vehicle driving related data, and a standardized data set H is constructed according to the preprocessed lamp control period data and vehicle driving related data, wherein the data preprocessing specific steps include data synchronization, abnormal value processing and data normalization;

[0020] Data synchronization refers to setting a uniform sliding time window ΔT, segmenting and aggregating data according to the time granularity of the uniform sliding time window ΔT, and selecting the median of the data in each time period as the representative value of the time period;

[0021] Abnormal value processing refers to identifying abnormal values using the IQR method and removing abnormal values;

[0022] Data normalization refers to normalizing the data using the Z-score standardization method.

[0023] Preferably, the S2 specific steps include:

[0024] S21, according to the disturbance area map Ga, identify the traffic driving path in the active area, divide each traffic driving path into a plurality of traffic sub-sections with the same length, and for each traffic sub-section, extract the vehicle driving trajectory of the traffic sub-section according to the standardized data set H, to obtain the trajectory point coordinates of the vehicle in the traffic sub-section, and calculate the direction vector of the vehicle in the traffic sub-section using the unit direction vector calculation formula and according to the direction vector of the vehicle in the traffic sub-section calculate the direction deviation angle a of each traffic sub-section i , wherein the direction deviation angle a i The calculation formula is:

[0025]

[0026] In the formula, a i represents the direction deviation angle of the ith traffic sub-section, cos -1 represents the inverse cosine function, represents the direction vector of the vehicle in the ith traffic sub-section before the traffic disturbance activity occurs, represents the direction vector of the vehicle in the ith traffic sub-section after the traffic disturbance activity occurs, and respectively represent the length of the direction vector of the vehicle in the ith traffic sub-section before and after the traffic disturbance activity occurs.

[0027] Preferably, the specific steps of S2 further comprise:

[0028] S22, according to the direction deviation angle a i of the traffic sub-section, for each traffic driving path, the path length change of the traffic driving path before and after the traffic disturbance activity is coupled with the direction deviation to model the path disturbance offset coefficient Py of the traffic driving path, to quantify the disturbance state of the traffic path, wherein the path disturbance offset coefficient Py calculation formula is:

[0029]

[0030] In the formula, P represents the total number of traffic sub-sections, represents the vehicle driving distance of the ith traffic sub-section after the traffic disturbance activity occurs, represents the vehicle driving distance of the ith traffic sub-section before the traffic disturbance activity occurs, cos represents the cosine function, and a i represents the direction deviation angle of the ith traffic sub-section.

[0031] Preferably, the specific steps of S2 further comprise:

[0032] S23, collect vehicle driving related data of the vehicle in the active area in the normal driving state as sample data to calculate the path disturbance offset coefficient Py, and perform statistical analysis, select the 90th percentile path disturbance offset coefficient Py as the path disturbance threshold Pyyz, and compare the maximum path disturbance offset coefficient Pymax with the path disturbance offset coefficient Py for the traffic driving path in the active area to evaluate the traffic state of the active area, and the specific evaluation content is as follows:

[0033] If the maximum path disturbance offset coefficient Pymax is greater than or equal to the path disturbance threshold Pyyz, it is determined that the traffic state of the active area is in an abnormal state, and at this time signal light cycle rhythm analysis needs to be performed.

[0034] If the maximum path disturbance offset coefficient Pymax is less than the path disturbance threshold Pyyz, it is determined that the traffic state of the active area is in a normal state, and at this time no processing is required.

[0035] Preferably, S3 includes the following specific steps:

[0036] S31, when the traffic state of the active area is in an abnormal state, identifying the middle time point of the green light termination time Tend and the green light starting time Tstart of each traffic signal light as the rhythm center point Tmid according to the light control cycle data;

[0037] S32, setting each two adjacent traffic signal lights as a traffic signal pair, for the traffic signal pair, calculating the rhythm center point difference ATmid of the two traffic signal lights in the same signal control cycle, and dividing the rhythm center point difference ATmid by the response time difference AT to obtain the phase change rate of the traffic signal pair, and calculating the average phase change rate to construct the phase dislocation propagation index Xw, wherein the response time difference AT refers to the rhythm response time difference between the traffic signal lights after the occurrence of a traffic disturbance event, and the rhythm response time refers to the time taken by a single traffic signal light to complete rhythm adjustment after the occurrence of a traffic disturbance event, which is obtained through traffic signal control log collection.

[0038] Preferably, S4 includes the following specific steps:

[0039] S41, combining the exit points and the entrance points in the active area according to the disturbance area map Ga and the GIS network to construct OD pairs, and performing the K shortest path algorithm on each pair of OD pairs to identify the total number M of traffic driving paths in the active area, and further calculating the average path disturbance offset coefficient Pyavg of the traffic driving paths in the active area, and performing dimensionless processing on the average path disturbance offset coefficient Pyavg and the phase dislocation propagation index Xw, and then performing coupling calculation to obtain the comprehensive rhythm adaptation index Zh, wherein the calculation formula of the comprehensive rhythm adaptation index Zh is:

[0040]

[0041] In the formula, ln represents a natural logarithm function, Pyavg represents a path disturbance offset coefficient mean value of a traffic driving path in an active area, exp represents an exponential function, and Xw represents a phase dislocation propagation exponent;

[0042] S42, obtain historical light control cycle data and vehicle driving related data of the active area through a traffic data storage center as sample data to calculate a comprehensive rhythm adaptation index Zh, and select the 10th percentile and the 80th percentile of the comprehensive rhythm adaptation index Zh as a first rhythm adaptation evaluation threshold Zhyz1 and a second rhythm adaptation evaluation threshold Zhyz2, and compare the first rhythm adaptation evaluation threshold Zhyz1 and the second rhythm adaptation evaluation threshold Zhyz2 with the comprehensive rhythm adaptation index Zh to evaluate the running state of the traffic signal control system and generate a system adjustment instruction. The specific evaluation content is as follows:

[0043] If the comprehensive rhythm adaptation index Zh is greater than or equal to the second rhythm adaptation evaluation threshold Zhyz2, it is determined that the running state of the traffic signal control system is in a third risk state, at this time the running state of the traffic signal control system is normal, and a third system adjustment instruction is generated;

[0044] If the first rhythm adaptation evaluation threshold Zhyz1 is less than the comprehensive rhythm adaptation index Zh and the comprehensive rhythm adaptation index Zh is less than the second rhythm adaptation evaluation threshold Zhyz2, it is determined that the running state of the traffic signal control system is in a second risk state, at this time the traffic signal control system has local dislocation, and a second system adjustment instruction is generated;

[0045] If the comprehensive rhythm adaptation index Zh is less than or equal to the first rhythm adaptation evaluation threshold Zhyz1, it is determined that the running state of the traffic signal control system is in a first risk state, at this time the running state of the traffic signal control system is unbalanced, and a first system adjustment instruction is generated.

[0046] Preferably, S5 specifically includes the following steps:

[0047] S51, according to the running state evaluation result of the traffic signal control system, use the system adjustment instruction to intelligently adjust the traffic signal control system, wherein the system adjustment instruction includes the third system adjustment instruction, the second system adjustment instruction and the first system adjustment instruction;

[0048] The third system adjustment instruction means maintaining the current traffic signal control system running state and adjusting the traffic signal control system data monitoring frequency to 150% of the standard data monitoring frequency;

[0049] The second system adjustment instruction refers to signal control cycle adjustment, wherein the signal control cycle adjustment refers to presetting a phase change threshold, comparing the phase change rate with the phase change threshold for the traffic signal pair in the active area, if the phase change rate is greater than or equal to the phase change threshold, identifying the traffic signal pair as a misaligned signal pair, postponing the green light time of the second traffic signal lamp in the misaligned signal pair by 3s, and extending the green light time window from the center time point to both sides by 2s;

[0050] The first system adjustment instruction refers to signal control cycle reconstruction, wherein the signal control cycle reconstruction refers to constructing a propagation path on the traffic topology for the identified misaligned signal pair, forming a graph structure Gz with a direction, Gz=(V, E), V represents a traffic intersection where the misaligned signal pair exists, E represents adjacent traffic intersections of the misaligned signal pair, and collecting feature data of each traffic signal lamp node in the past N signal control cycles, including signal control cycle length, green light offset, green light average vehicle speed, and green light end queue residual amount, to construct a feature vector F, and based on the graph structure Gz and the feature vector F, using a graph neural network for inference training, and taking traffic queue length, rhythm offset, and vehicle average travel delay as adaptive constraint parameters, constructing a function model for rhythm optimization learning, outputting the optimal signal control cycle of each traffic signal lamp node, to generate traffic signal lamp control instructions for intelligent control of traffic signal lamps.

[0051] The intelligent city traffic planning system based on deep learning includes a data collection module, a traffic state analysis module, a cycle rhythm analysis module, a comprehensive evaluation module, and an intelligent adjustment module.

[0052] The data collection module is used to determine the activity area of the traffic disturbance activity through the city social media platform, to construct a disturbance area graph Ga, and to collect vehicle driving related data and lamp control cycle data in the disturbance area graph Ga, to construct a standardized data set H.

[0053] The traffic state analysis module is used to identify the traffic driving path according to the disturbance area graph Ga, and to analyze the path length change and direction offset of the vehicle before and after the traffic disturbance activity, to evaluate the traffic state in the activity area, and to analyze the signal cycle rhythm when the traffic state in the activity area is in an abnormal state.

[0054] The cycle rhythm analysis module is used to calculate the rhythm center point difference of the traffic signal lamp according to the lamp control cycle data when the traffic state in the activity area is in an abnormal state, to quantify the rhythm consistency and response coordination between the traffic signal lamps.

[0055] The comprehensive evaluation module is used for identifying the total number M of traffic driving paths according to the disturbance area map Ga using the K shortest path algorithm, analyzing the operation state of the traffic signal control system in the activity area, classifying the risk level, and generating system adjustment instructions of different risk levels;

[0056] The intelligent adjustment module is used for intelligently adjusting the traffic signal control system according to the system adjustment instructions.

[0057] The application provides a smart city traffic planning method and system based on deep learning.

[0058] (1) By constructing the disturbance area map Ga, collecting vehicle driving related data and light control cycle data in the disturbance area map Ga, the traffic disturbance area caused by the city emergency event is accurately modeled and data is collected, then the traffic path is divided into multiple sub-sections, the path disturbance offset coefficient Py is calculated, and the quantitative analysis of the direction offset and length change of the vehicle path is realized. This mechanism effectively overcomes the problems of rough model and local failure in the traditional path analysis method which takes the main road as the center and ignores the disturbance of the activity area. It can realize dynamic perception and abnormal judgment of the traffic state in the activity area, and provide real-time basis for subsequent signal system adjustment.

[0059] (2) On the basis of identifying the traffic abnormal state, the cycle rhythm of the traffic signal is further analyzed, a signal phase change rate model based on the rhythm center point Tmid is constructed, a phase misalignment propagation index Xw is constructed combined with the response time difference Δτ, and the comprehensive rhythm adaptation index Zh is constructed by coupling the phase misalignment propagation index Xw and the path disturbance offset coefficient Pyavg. Through the threshold value Zhyz1 and Zhyz2 set by the quantile, the multi-level risk judgment and classified adjustment of the operation state of the signal control system are realized. This method breaks through the problems of lack of rhythm misalignment propagation chain analysis and lack of dynamic classification strategy in the prior art, and can more carefully identify the rhythm imbalance state under different mismatch degrees, thereby improving the resolution accuracy and intervention timeliness of the system to the rhythm abnormality.

[0060] (3) By constructing the propagation path graph Gz containing the misaligned signal pair, collecting the historical signal control cycle characteristics to form the feature vector F, and combining the graph neural network inference model, the optimal control cycle strategy is learned and generated based on multiple adaptive constraint parameters such as traffic queue length, rhythm offset and vehicle traffic delay, and the signal control instructions for specific intersections are formed. This mechanism significantly improves the intelligent adjustment ability of the signal control system in the dynamic disturbance scene, overcomes the problem of relying on fixed templates or artificial experience in traditional cycle adjustment, realizes closed-loop adaptive control from path disturbance recognition to cycle reconstruction, and provides a replicable, scalable and generalizable system solution for smart city traffic governance. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 A flowchart of the traffic planning method based on deep learning of the smart city of the present application;

[0062] Figure 2 A flowchart of the traffic planning system based on deep learning of the smart city of the present application;

[0063] Figure 3 A schematic diagram of the standardized data set H of the present application;

[0064] Figure 4 A schematic diagram of the running state evaluation result of the traffic signal control system of the present application. DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0066] Embodiment 1

[0067] Please refer to Figure 1 The present application provides a traffic planning method based on deep learning of a smart city. To achieve the above purpose, the present application is implemented by the following technical solutions: comprising the following steps:

[0068] S1, determining the activity area of the traffic disturbance activity through the city social media platform to construct a disturbance area map Ga, and collecting vehicle driving related data and light control cycle data in the disturbance area map Ga to construct a standardized data set H;

[0069] S2, identifying the traffic driving path according to the disturbance area map Ga, and analyzing the path length change and direction deviation of the vehicle before and after the start of the traffic disturbance activity to evaluate the traffic state of the activity area, and when the traffic state of the activity area is in an abnormal state, analyzing the signal light cycle rhythm;

[0070] S3, when the traffic state of the activity area is in an abnormal state, calculating the rhythm center point difference of the traffic signal lights according to the light control cycle data to quantify the rhythm consistency and response coordination between the traffic signal lights;

[0071] S4, identifying the total number M of traffic driving paths using the K shortest path algorithm according to the disturbance area map Ga, analyzing the running state of the traffic signal control system in the activity area, classifying the risk levels, and generating system adjustment instructions of different risk levels;

[0072] S5, intelligently adjusting the traffic signal control system according to the system adjustment instruction.

[0073] In this embodiment, by constructing the disturbance region map Ga and combining deep learning means, the local traffic anomaly caused by the sudden activity in the smart city is dynamically modeled and responded, which significantly improves the adaptive planning ability of the urban traffic system in complex scenarios. Compared with the existing technology based on static traffic model or linear main guide path strategy, the disturbance activity area is automatically identified from the government system, and the actual collected vehicle path data is standardized processed, ensuring the timeliness and accuracy of the data, solving the problem of slow response of traditional methods to temporary events, and through the coupling modeling of the path length change rate and the direction deviation angle of the traffic path before and after the disturbance, the influence of the disturbance on the actual traffic path can be accurately quantified, providing a reliable basis for abnormal state identification. At the same time, the rhythm analysis mechanism is introduced to dynamically monitor the traffic signal control cycle, effectively identifying the rhythm misalignment risk. Further, the application realizes the operation state judgment of the traffic signal control system in the whole disturbance area through the global path analysis and system evaluation strategy, and generates adjustment instructions according to the grading results, realizing the closed-loop response from abnormal identification to linkage control. Compared with the traditional passive adjustment mechanism, the intelligent adjustment framework constructed by the application has the advantages of high adaptability, low delay and scalability, and can be widely applied to complex scenarios such as urban marathons, large concerts, exhibition openings, etc. It solves the problems of slow identification, rough control and delayed recovery of the existing traffic system, and improves the intelligent, accurate and real-time level of smart city traffic planning.

[0074] Embodiment 2

[0075] This embodiment is an explanation and description in embodiment 1, please refer to Figure 1 and Figure 2 , specifically: S1 specific steps include:

[0076] S11, according to the crawler grabbing tool Scrapy, grabbing the traffic disturbance activity related fields on the city social media platform, including text content, publishing time and location information, to identify the traffic disturbance activity in the city, and combining the city map database, extracting the boundary point coordinates of the traffic disturbance activity, and according to the boundary point coordinates of the traffic disturbance activity, using the convex hull algorithm, sorting all boundary point coordinates by polar angle, starting from the leftmost boundary point, connecting in the counterclockwise direction in turn, forming a closed figure region, that is, the activity area of the traffic disturbance activity, wherein the boundary point coordinates refer to the GPS geographical position point coordinates used to describe the boundary of the traffic disturbance activity influence area, which are obtained by analyzing the location information of the traffic disturbance activity.

[0077] Convex hull algorithm Convex hull algorithm is an algorithm for calculating the minimum convex polygon of a given point set on a plane, and the convex hull can be understood as the convex polygon formed by connecting the outermost points in the point set.

[0078] Traffic disturbance activities refer to a set of events that cause significant disturbance to the traffic operation state of existing urban roads, induce changes in vehicle operation paths, or adjust signal timing, including but not limited to large-scale mass events, emergency road closures, traffic accidents-induced road congestion, and urban disasters, etc.

[0079] Urban map database refers to a data set containing road structure, traffic nodes, infrastructure layout, and their geometric attributes and spatial location relationships within a city, usually provided by city traffic management departments or third-party surveying and mapping agencies, supporting GIS format storage and standardized query.

[0080] S12, in the activity area of the traffic disturbance activity, set each traffic signal as a traffic signal node, and traverse the entire activity area of the traffic disturbance activity, construct a traffic signal node set, and call the signal control platform log to obtain the light control period data of the traffic signal node, wherein the light control period data includes the green light start time Tstart and the green light end time Tend, and according to the urban map database, obtain the GPS coordinate data of each traffic signal node, to construct a disturbance area graph Ga, wherein the disturbance area graph Ga=(Na, Ea), Na represents the traffic signal node set, and Ea represents the road connection edge set;

[0081] Signal control platform log refers to the signal control operation record log deployed in the background platform of the traffic signal control system, which is a time series data set reflecting the running state of the traffic signal and the change history of the control strategy;

[0082] S13, a plurality of data collection intervals are set in the activity area, and radar speed measurement equipment and high-definition card slot cameras are deployed in each data collection interval to obtain real-time vehicle driving related data in the activity area, wherein the vehicle driving related data includes vehicle driving trajectory and vehicle driving distance d;

[0083] Vehicle driving trajectory refers to the driving route of a vehicle on a traffic driving road, including the trajectory point coordinates of the vehicle on the traffic driving road, which is obtained by recognizing and acquiring a high-definition card slot camera;

[0084] Vehicle driving distance d refers to the actual trajectory length of a vehicle moving along the traffic driving route in the activity area, which is obtained by integrating the speed of the radar speed measurement equipment at each coordinate point, wherein the radar speed measurement equipment is installed on the roadside signal pole, monitoring pole and power pole by fixed bracket.

[0085] S1 specific steps also include:

[0086] S14, data preprocessing is performed on the collected lamp control cycle data and vehicle driving related data, and a standardized data set H is constructed according to the preprocessed lamp control cycle data and vehicle driving related data, wherein the specific steps of data preprocessing include data synchronization, abnormal value processing and data normalization;

[0087] Data synchronization refers to setting a unified sliding time window ΔT, segmenting and aggregating the data according to the time granularity of the unified sliding time window ΔT, and selecting the median of the data in each time period as the representative value of the time period;

[0088] Abnormal value processing refers to identifying abnormal values using the IQR method and removing abnormal values;

[0089] Data normalization refers to normalizing the data using the Z-score standardization method.

[0090] In this embodiment, by introducing the boundary point set {Va}, the actual activity area of the traffic disturbance event is determined, and all traffic signal nodes in this area are included in the modeling category, breaking through the limitations of traditional selection based on trunk roads or fixed nodes, realizing comprehensive perception and spatial reconstruction of complex disturbance areas, and combining the signal control platform log and the city map database, the control cycle and spatial position information of the traffic signal are accurately obtained, ensuring that the modeling result has spatiotemporal consistency, effectively improving the decision basis for subsequent signal rhythm analysis and linkage control. At the same time, this embodiment uses radar speed measurement equipment and high-definition card slot cameras for data collection, not only improving the collection accuracy of path turning angle and driving distance data, but also enhancing the ability to describe the real behavior of vehicles through multi-source fusion. In view of the data reliability problem, multiple preprocessing strategies such as unified sliding time window ΔT, IQR abnormal value removal and Z-score normalization are introduced, a high-quality, comparable and low-noise data standard set H is established, and high-quality input samples are provided for the deep learning model. Overall, this embodiment effectively solves the problems of boundary ambiguity, arbitrary node selection, data redundancy and noise in traditional traffic disturbance modeling, so that the system has stronger spatiotemporal recognition ability, behavior modeling ability and response preposition ability when facing sudden events, providing a solid data and structural support foundation for the intelligent city traffic system.

[0091] Embodiment 3

[0092] This embodiment is an explanation and description in embodiment 2, please refer to Figure 1 , in particular:

[0093] S2 includes the following specific steps:

[0094] S21, according to the disturbance area map Ga, identify the traffic driving path in the active area, divide each traffic driving path into several traffic sub-sections with the same length, and for each traffic sub-section, extract the vehicle driving trajectory of the traffic sub-section according to the standardized data set H, to obtain the trajectory point coordinates of the vehicle in the traffic sub-section, and calculate the direction vector of the vehicle in the traffic sub-section using the unit direction vector calculation formula And according to the direction vector of the vehicle in the traffic sub-section Calculate the direction deviation angle a of each traffic sub-section i Where the direction deviation angle a i The calculation formula is:

[0095]

[0096] In the formula, a i represents the direction deviation angle of the i-th traffic sub-section, cos -1 represents the inverse cosine function, represents the direction vector of the vehicle in the i-th traffic sub-section before the traffic disturbance activity occurs, represents the direction vector of the vehicle in the i-th traffic sub-section after the traffic disturbance activity occurs, And respectively represent the length of the direction vector of the vehicle in the i-th traffic sub-section before and after the traffic disturbance activity occurs.

[0097] Formula derivation process and physical meaning: the formula is derived from the definition of the angle between two vectors in three-dimensional Euclidean space, which is a vector angle calculation formula. It is widely used in trajectory analysis, automatic driving trajectory planning, path deviation identification, video pose identification, and graphics. The calculation of the direction deviation angle can quantify the direction deviation effect caused by the path disturbance. If the vehicle deviates from the original direction for detour, such as changing from a straight line to a turn, the direction deviation angle will increase;

[0098] Specific examples are as follows:

[0099] Suppose that due to the marathon activity in a certain urban area, some main road sections are closed, and a traffic driving path in a traffic sub-section before the traffic disturbance activity occurs from intersection A to B, with a length of 100m and a northeast direction. The direction vector of the vehicle before the traffic disturbance activity occurs is (100, 0), and the direction vector of the vehicle after the traffic disturbance activity occurs is (71, 71). At this time, the calculation of the direction deviation angle is 45°.

[0100] S2 specific steps also include:

[0101] ​​S22, according to the direction offset angle a of the traffic sub-section i For each traffic travel path, the path length change and the direction offset of the traffic travel path before and after the traffic disturbance activity are coupled to model the path disturbance offset coefficient Py of the traffic travel path, which quantifies the disturbance state of the traffic path, wherein the calculation formula of the path disturbance offset coefficient Py is:

[0102]

[0103] In the formula, P represents the total number of traffic sub-sections, represents the vehicle travel distance of the ith traffic sub-section after the traffic disturbance activity, represents the vehicle travel distance of the ith traffic sub-section before the traffic disturbance activity, and cos represents the cosine function, a i represents the direction offset angle of the ith traffic sub-section.

[0104] Formula derivation process and physical meaning: in the formula, the path length change rate is the same path increment on the ith traffic field, cos(a i ) represents the projection factor of the direction offset, which projects the distance change onto the main travel path to evaluate the disturbance intensity in the main direction, and then multiplies the path length change rate and the projection factor of the direction offset, which is the length change multiplied by the direction offset, to form a disturbance intensity quantization expression under the direction constraint. Finally, the disturbance intensities of all traffic sub-sections are combined to build a path-level disturbance index, which represents the disturbance degree of the entire traffic travel path, considering the spatial detour change and the direction offset.

[0105] Specific examples are as follows:

[0106] Suppose a traffic travel path contains 5 traffic sub-sections, and the direction offset angles of the traffic sub-sections are 0°, 30°, 60°, 15° and 45°, respectively. The angles are converted to cosine values, which are 1.000, 0.866, 0.500, 0.966 and 0.707, respectively. The vehicle travel distances before the traffic disturbance activity are 100, 80, 60, 120 and 90, respectively. The vehicle travel distances after the traffic disturbance activity are 100, 96, 75, 108 and 117, respectively. At this time, the path disturbance offset coefficient Py of the traffic travel path is calculated as 0.0827, i.e. the average offset intensity of the traffic travel path after the traffic disturbance activity is 8.3%.

[0107] The specific steps of S2 further include:

[0108] S23, collect vehicle driving related data of the vehicle in the active area in the normal driving state as sample data to calculate the path disturbance offset coefficient Py, and perform statistical analysis, select the 90th percentile of the path disturbance offset coefficient Py as the path disturbance threshold Pyyz, and compare the maximum path disturbance offset coefficient Pymax with the path disturbance offset coefficient Py for the traffic driving path in the active area to evaluate the traffic state of the active area, and the specific evaluation content is as follows:

[0109] If the maximum path disturbance offset coefficient Pymax is greater than or equal to the path disturbance threshold Pyyz, it is determined that the traffic state of the active area is in an abnormal state, and at this time signal light cycle rhythm analysis needs to be performed.

[0110] If the maximum path disturbance offset coefficient Pymax is less than the path disturbance threshold Pyyz, it is determined that the traffic state of the active area is in a normal state, and at this time no processing is needed.

[0111] In this embodiment, a coupling modeling mechanism based on the fusion of path length change and direction offset angle double characteristics is proposed, which effectively improves the sensitivity and reliability of the traffic state abnormality recognition. By subdividing the traffic driving path into several equal-length traffic sub-sections and calculating the direction offset angle of the sub-sections before and after the disturbance, the microscopic path offset trend caused by vehicle behaviors such as avoidance and detour can be accurately captured, thereby avoiding the recognition lag or misjudgment caused by the coarse granularity of the traditional path comparison method. Combined with the path length change, the path disturbance offset coefficient Py is established, and the multi-dimensional and dynamic feature recognition of traffic disturbance is further realized, which is helpful for accurate modeling of vehicle behavior under complex disturbance background. At the same time, this embodiment introduces the 90th percentile to establish a dynamic path disturbance evaluation threshold, effectively eliminating the interference of extreme data on the judgment result, making the traffic state determination have strong robustness and generalization. The path disturbance offset coefficient Py not only serves as an important criterion for whether the signal light cycle rhythm analysis needs to be started, but also provides a quantifiable triggering mechanism for subsequent rhythm analysis and cycle adjustment, breaking through the single-dimensional idea of judging congestion and abnormality only by average speed or travel time in the existing method, realizing the pre-quantitative recognition of the traffic state under sudden traffic disturbance events, providing a more scientific, stable and efficient response entry for smart city traffic regulation, and significantly improving the active governance ability of the system.

[0112] Embodiment 4

[0113] This embodiment is an explanation and description in Embodiment 3, please refer to Figure 1 , in detail: the specific steps of S3 include:

[0114] S31, S31, when the passing state of the active area is in an abnormal state, identifying the middle time point of the green light end time Tend and the green light start time Tstart of each traffic signal lamp as the rhythm center point Tmid according to the lamp control cycle data;

[0115] S32, setting every two adjacent traffic signal lamps as a traffic signal pair, calculating the rhythm center point difference ATmid of the two traffic signal lamps in the same signal control cycle for the traffic signal pair, dividing the rhythm center point difference ATmid by the response time difference AT to obtain the phase change rate of the traffic signal pair, and calculating the average of the phase change rate to construct the phase misalignment propagation index Xw, wherein the response time difference AT refers to the rhythm response time difference between the traffic signal lamps after the occurrence of a traffic disturbance event, and the rhythm response time refers to the time taken by a single traffic signal lamp to complete rhythm adjustment after the occurrence of a traffic disturbance event, which is obtained through traffic signal control log collection.

[0116] The phase misalignment propagation index Xw represents the diffusion intensity of the phase offset of adjacent traffic signal lamp nodes in the traffic signal network under the influence of disturbance in a unified signal control cycle;

[0117] In this embodiment, a quantitative analysis method for the rhythm response of traffic signal lamps is proposed, and the rhythm characteristic parameter "phase misalignment propagation index Xw" is constructed to evaluate the rhythm consistency and response coordination of the traffic signal control system under the background of traffic disturbance. By setting the signal control cycle center point Tmid as the rhythm analysis reference, the interference caused by the differences in red, yellow and green phase lengths can be effectively avoided, so that the core rhythm of the signal lamp timing setting can be more truly reflected. Further, by using the ratio between the time difference of the rhythm center points of the traffic signal lamps and the response time difference, the phase change rate is constructed and the average is calculated, which can reveal the propagation law of the rhythm offset caused by temporary adjustment of the control strategy in the signal network. This breaks through the evaluation logic of traditional traffic signal control evaluation which only relies on green light timing coverage or cycle minimization, introduces a system-level rhythm propagation perspective, and can be used to identify the rhythm mismatch chain and coupling error diffusion path caused by traffic disturbance, thereby providing quantifiable risk basis and path identification support for subsequent rhythm regulation strategy formulation. In addition, the system's detection ability for rhythm-level abnormalities is enhanced, the prevention and control ability of the urban traffic signal control system for network-level cascading disorder risks under sudden events is improved, and the linkage understanding ability and regulation foresight of the intelligent transportation system for "signal lamp group behavior" are strengthened.

[0118] Embodiment 5

[0119] Please refer to Figure 1 and Figure 4 , specifically:

[0120] S4 Specific steps include:

[0121] S41, according to the disturbance area map Ga and GIS network, the exit point and the entrance point in the active area are combined to construct the OD pair, and the K shortest path algorithm is executed for each OD pair to identify the total number of traffic travel paths M in the active area, and further calculate the path disturbance offset coefficient average Pyavg of the traffic travel path in the active area, and the path disturbance offset coefficient average Pyavg and the phase misalignment propagation index Xw are dimensionless, and then coupled to obtain the comprehensive rhythm adaptation index Zh, wherein the comprehensive rhythm adaptation index Zh calculation formula is:

[0122]

[0123] In the formula, ln represents the natural logarithm function, Pyavg represents the path disturbance offset coefficient average of the traffic travel path in the active area, exp represents the exponential function, and Xw represents the phase misalignment propagation index.

[0124] The formula derivation process and physical meaning: the formula is essentially a nonlinear combination function, which uses dimensionless processed characteristic indicators for coupling calculation, combining two-dimensional characteristic indicators, including path disturbance offset strength and rhythm misalignment transmission ability, wherein the natural logarithm function ln is used because the disturbance increases the adjustment demand, which is not linear but decreasing, with marginal weakening effect, +1 is to prevent negative infinity when the disturbance is 0, and exp is the exponential function, which maps the misalignment propagation ability to the scalability function of the adjustment mechanism, and finally performs coupling calculation to form a comprehensive index, which represents whether the current acquisition area has the ability to effectively reconstruct the signal control rhythm of the traffic system. The larger the Zh value is, the better the system can adapt to the traffic changes caused by path disturbance and perform corresponding signal rhythm buffering.

[0125] GIS network is a spatial connection network model with topological relationship based on geographic spatial data, which is used to describe the spatial and logical relationship between different nodes and edges in the actual road system;

[0126] K shortest path algorithm is used to find the first K paths in the traffic network from the starting point to the ending point, which is used to support the construction of path disturbance offset coefficient and subsequent traffic rhythm reconstruction analysis;

[0127] The specific example is as follows:

[0128] Assuming that the path disturbance offset coefficient average of the traffic travel path in the active area is 0.0877, and the phase misalignment propagation index is 1.4, the comprehensive rhythm adaptation index Zh is calculated as 0.091.

[0129] S42, through the traffic data storage center, obtain the historical light control period data and vehicle driving related data of the activity area as sample data to calculate the comprehensive rhythm adaptation index Zh, and select the 10th percentile and the 80th percentile of the comprehensive rhythm adaptation index Zh as the first rhythm adaptation evaluation threshold Zhyz1 and the second rhythm adaptation evaluation threshold Zhyz2, and compare the first rhythm adaptation evaluation threshold Zhyz1 and the second rhythm adaptation evaluation threshold Zhyz2 with the comprehensive rhythm adaptation index Zh to evaluate the running state of the traffic signal control system, and generate system adjustment instructions, and the specific evaluation content is as follows:

[0130] If the comprehensive rhythm adaptation index Zh is greater than the second rhythm adaptation evaluation threshold Zhyz2, it is determined that the running state of the traffic signal control system is in the third risk state, at this time the running state of the traffic signal control system is normal, and the third system adjustment instruction is generated;

[0131] If the first rhythm adaptation evaluation threshold Zhyz1 is less than the comprehensive rhythm adaptation index Zh and the comprehensive rhythm adaptation index Zh is less than the second rhythm adaptation evaluation threshold Zhyz2, it is determined that the running state of the traffic signal control system is in the second risk state, at this time the traffic signal control system exists local misplacement, and the second system adjustment instruction is generated;

[0132] If the comprehensive rhythm adaptation index Zh is less than or equal to the first rhythm adaptation evaluation threshold Zhyz1, it is determined that the running state of the traffic signal control system is in the first risk state, at this time the running state of the traffic signal control system is unbalanced, and the first system adjustment instruction is generated.

[0133] In this embodiment, by calling the GIS road network, the system automatically constructs the OD pair of the entry point and the exit point in the activity area, and combines the K shortest path algorithm to systematically identify the traffic path topological structure and the path number, effectively avoiding ignoring the important path distribution of the local perturbation region by only taking the trunk road as the analysis object, further, the system will all path disturbance states be averaged and coupled with the phase misplacement propagation index Xw to model, realize the linkage evaluation of "path disturbance behavior" and "signal rhythm misplacement", effectively bridge the gap between "vehicle behavior analysis" and "signal response analysis" in traditional traffic evaluation, at the same time, based on the historical data, a multi-stage rhythm adaptation threshold distribution is constructed to divide the risk level, and the dynamic triggering of the fine regulation strategy is realized. Compared with the static optimization mode of the existing signal control strategy based on traffic volume or local timing adjustment, the present scheme has stronger adaptability and response foresight, can identify the potential linkage imbalance risk of the signal system in advance under the disturbance situation of urban large-scale activities, sudden events or structural congestion, and provide hierarchical regulation instructions according to the risk level, introduce the rhythm adaptation ability of "self-sensing-self-evaluation-self-regulation" for the traffic signal control system, and improve the resilience and intelligent level of the overall urban traffic operation.

[0134] Embodiment 6

[0135] This embodiment is an explanation and illustration in embodiment 2, please refer to Figure 1 , specifically: S5 specific steps include:

[0136] S51, according to the running state evaluation result of the traffic signal control system, using system adjustment instruction to intelligently adjust the traffic signal control system, wherein the system adjustment instruction includes third system adjustment instruction, second system adjustment instruction and first system adjustment instruction;

[0137] The third system adjustment instruction refers to maintaining the current traffic signal control system running state and adjusting the traffic signal control system data monitoring frequency to 150% of the standard data monitoring frequency;

[0138] The second system adjustment instruction refers to the misaligned signal control cycle adjustment, wherein the signal control cycle adjustment is to preset the phase change threshold, and for the traffic signal pair in the active area, the phase change rate is compared and analyzed with the phase change threshold, if the phase change rate ≥ the phase change threshold, the traffic signal pair is identified as misaligned signal pair, the green light time of the second traffic signal lamp in the misaligned signal pair is pushed back 3s, and the green light time window is extended by 2s from the center time point to both sides;

[0139] The first system adjustment instruction refers to signal control cycle reconstruction, wherein the signal control cycle reconstruction refers to constructing the propagation path on the traffic topology for the identified misaligned signal pair, forming a graph structure Gz with direction, Gz=(V,E), V represents the traffic intersection where the misaligned signal pair exists, E represents the adjacent traffic intersection of the misaligned signal pair, and the feature data of each traffic signal lamp node in the past N signal control cycles is collected, including signal control cycle length, green light offset, green light average vehicle speed and green light end queue residual amount, to construct a feature vector F, and based on the graph structure Gz and the feature vector F, the graph neural network is used for inference training, and the traffic queue length, rhythm offset and vehicle average delay are used as adaptive constraint parameters to construct a function model for rhythm optimization learning, and the optimal signal control cycle of each traffic signal lamp node is output to generate traffic signal lamp control instruction to intelligently control the traffic signal lamp.

[0140] The signal control cycle length refers to the complete control cycle time of the traffic signal lamp composed of green light, yellow light and red light, which is obtained through the traffic signal controller log;

[0141] The green light offset refers to the time difference between the green light start time of the traffic signal lamp and the reference start time;

[0142] The green light passing vehicle speed average refers to the average driving speed of vehicles passing the intersection during the effective period of the green light, which is obtained by a radar speed measurement device;

[0143] The green light end queuing residual quantity refers to the number of vehicles waiting for traffic at the instant of the end of the green light, which is obtained by recognizing a high-definition turret camera;

[0144] The traffic queuing length refers to the total number of vehicles accumulated in front of the stop line during red light control, which is obtained by recognizing a high-definition turret camera;

[0145] The rhythm offset quantity refers to the offset time difference between the current signal timing rhythm and the ideal rhythm in the regional optimization control strategy, which is obtained by comparing the current signal timing table with the standard signal timing table;

[0146] The vehicle average passing delay refers to the average waiting time of vehicles at the intersection, which is obtained by recognizing the waiting time of each vehicle at the intersection using a high-definition turret camera and calculating the average;

[0147] The graph neural network is a neural network structure that takes a graph structure as input and models and learns the nodes or the whole graph through an information propagation mechanism between adjacent nodes. In the present scheme, the graph neural network outputs the optimal signal control period of each traffic signal light node for intelligent adjustment of the traffic signal light.

[0148] In the present embodiment, a hierarchical and intelligent rhythm adjustment strategy is constructed for the running state of the traffic signal control system under different risk levels. When facing the third risk level, i.e., the system linkage imbalance state, a graph neural network modeling and adaptive learning mechanism is introduced to effectively break through the passive response limitations of traditional signal timing adjustment methods. First, the system distinguishes three types of system adjustment instructions according to the evaluation results, realizing gradient control from the maintenance state to deep optimization. Among them, the second system adjustment instruction identifies misaligned signal pairs through the comparison of the phase change rate with the preset threshold and fine-tunes the green light timing, which has the characteristics of immediacy and low intervention and is suitable for the second risk level. In the first risk level state, the traffic topology graph structure is reconstructed, the propagation path of misaligned signal pairs in space is included in the unified modeling, and the graph structure and feature vector are constructed combining the node feature data of multiple historical periods, so as to realize rhythm optimization learning based on the graph neural network. By introducing the traffic queuing length, rhythm offset quantity, and vehicle average passing delay as constraint indicators, the model can output the optimal control period of each signal light node for the regional linkage imbalance problem, thereby realizing the reconstruction and recovery of the rhythm linkage in the whole region. This method realizes the reconstruction of the control period driven by the whole region and data while taking into account the historical running situation and real-time disturbance response, significantly improves the adaptive adjustment ability of the signal system to complex traffic disturbances, and effectively enhances the intelligent resilience and operational stability of the urban traffic management system in extreme situations.

[0149] Example 7

[0150] See Figure 1 and Figure 2 Figure 2 The deep learning-based smart city traffic planning system includes a data collection module, a traffic state analysis module, a cycle rhythm analysis module, a comprehensive evaluation module, and an intelligent adjustment module.

[0151] The data collection module is used to determine the activity area of the traffic disturbance activity through the city social media platform, to construct a disturbance area map Ga, and to collect vehicle driving related data and light control cycle data within the disturbance area map Ga, to construct a standardized data set H.

[0152] The traffic state analysis module is used to identify the traffic driving path according to the disturbance area map Ga, and to analyze the path length change and direction deviation of the vehicle before and after the start of the traffic disturbance activity, to evaluate the traffic state of the activity area, and to perform signal light cycle rhythm analysis when the traffic state of the activity area is in an abnormal state.

[0153] The cycle rhythm analysis module is used to calculate the rhythm center point difference of the traffic signal lights according to the light control cycle data when the traffic state of the activity area is in an abnormal state, to quantify the rhythm consistency and response coordination between the traffic signal lights.

[0154] The comprehensive evaluation module is used to identify the total number of traffic driving paths M using the K shortest path algorithm according to the disturbance area map Ga, to analyze the running state of the traffic signal control system in the activity area, to perform risk level division, and to generate system adjustment instructions for different risk levels.

[0155] The intelligent adjustment module is used to intelligently adjust the traffic signal control system according to the system adjustment instructions.

[0156] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the application.

Claims

1. A deep learning-based smart city traffic planning method, characterized by: Includes the following steps: S1. Through urban social media platforms, determine the activity area of ​​traffic disturbance activities to construct a disturbance area map Ga, and collect vehicle driving-related data and traffic light control cycle data within the disturbance area map Ga to construct a standardized data set H; S2. Based on the disturbance area map Ga, identify traffic travel paths and analyze the changes in path length and direction deviation of vehicles before and after the start of traffic disturbance activities to assess the traffic status of the activity area. When the traffic status of the activity area is abnormal, conduct traffic light cycle rhythm analysis. S3. When the traffic conditions in the activity area are abnormal, calculate the difference in the rhythm center point of the traffic lights based on the light control cycle data, and quantify the rhythm consistency and response coordination between traffic lights. S4. Based on the disturbance area map Ga, use the K-shortest path algorithm to identify the total number of traffic travel paths M, analyze the operating status of the traffic signal control system within the activity area, classify the risk levels, and generate system adjustment instructions for different risk levels. S5. Based on the system adjustment instructions, intelligently adjust the traffic signal control system.

2. The deep learning-based smart city traffic planning method according to claim 1, characterized in that: The specific steps in S1 include: S11. Using the web crawling tool Scrapy, relevant fields of traffic disturbance activities are crawled from urban social media platforms, including text content, posting time and location information, to identify traffic disturbance activities in the city. Combined with the urban map database, the boundary point coordinates of traffic disturbance activities are extracted. Based on the boundary point coordinates of traffic disturbance activities, the convex hull algorithm is used to sort all boundary point coordinates by polar angle. Starting from the bottom left boundary point, the coordinates are connected in a counterclockwise direction to form a closed graphical region, which is the activity area of ​​traffic disturbance activities. The boundary point coordinates refer to the GPS geographic location coordinates used to characterize the boundary of the area affected by traffic disturbance activities, which are obtained by parsing the location information of traffic disturbance activities. S12. Within the activity area of ​​the traffic disturbance, each traffic light is defined as a traffic light node. The entire activity area of ​​the traffic disturbance is traversed to construct a set of traffic light nodes. The traffic control platform log is called to obtain the light control cycle data of the traffic light nodes. The light control cycle data includes the green light start time Tstart and the green light end time Tend. Based on the city map database, the GPS coordinate data of each traffic light node is obtained to construct the disturbance area map Ga. The disturbance area map Ga = (Na, Ea), where Na represents the set of traffic light nodes and Ea represents the set of road connecting edges. S13. Set up several data collection intervals within the activity area, and deploy radar speed measurement equipment and high-definition checkpoint cameras in each data collection interval to acquire vehicle driving-related data within the activity area in real time. Among them, vehicle driving-related data includes vehicle driving trajectory and vehicle driving distance d. Vehicle trajectory refers to the route a vehicle takes on a road, including the coordinates of the points on the road where the vehicle travels, which is obtained through high-definition checkpoint cameras. The vehicle travel distance d refers to the actual trajectory length of the vehicle along the traffic route within the activity area. It is obtained by collecting the vehicle's passage time and speed at each coordinate point through radar speed measurement equipment and performing speed integration calculation. The radar speed measurement equipment is installed on roadside traffic light poles, monitoring poles, and utility poles using fixed brackets.

3. The deep learning-based smart city traffic planning method according to claim 2, characterized in that: The specific steps in S1 also include: S14. Perform data preprocessing on the collected light control cycle data and vehicle driving related data, and construct a standardized data set H based on the preprocessed light control cycle data and vehicle driving related data. The specific steps of data preprocessing include data synchronization, outlier handling and data normalization. Data synchronization refers to setting a uniform sliding time window ΔT, segmenting and aggregating data according to the time granularity of the uniform sliding time window ΔT, and selecting the median of the data in each time period as the representative value of the time period. Outlier handling refers to using the IQR method to identify outliers and remove them. Data normalization refers to the process of normalizing data using the Z-score standardization method.

4. The deep learning-based smart city traffic planning method according to claim 3, characterized in that: The specific steps in S2 include: S21. Based on the disturbance area map Ga, identify traffic paths within the activity area, divide each traffic path into several traffic segments of equal length, and for each traffic segment, extract the vehicle trajectory based on the standardized data set H to obtain the coordinates of the vehicle's trajectory points within the traffic segment, and calculate the vehicle's direction vector within the traffic segment using the unit direction vector calculation formula. And based on the vehicle's direction vector in the traffic sub-segment Calculate the direction offset angle α for each traffic segment. i Wherein, the direction offset angle α i The calculation formula is: In the formula, α i Cos(i) represents the directional offset angle of the i-th traffic segment. -1 Represents the inverse cosine function. This represents the direction vector of a vehicle in the i-th traffic sub-segment before the traffic disturbance occurs. This represents the direction vector of a vehicle in the i-th traffic sub-segment after a traffic disturbance occurs. and Let represent the magnitudes of the direction vectors of vehicles in the i-th traffic sub-segment before and after the traffic disturbance, respectively.

5. The deep learning-based smart city traffic planning method according to claim 4, characterized in that: The specific steps in S2 also include: S22, Based on the direction offset angle α of the traffic segment i For each traffic path, the changes in path length and direction offset before and after traffic disturbance are coupled and modeled to obtain the path disturbance offset coefficient Py, which quantifies the disturbance state of the traffic path. The formula for calculating the path disturbance offset coefficient Py is as follows: In the formula, P represents the total number of traffic sub-segments. This represents the distance traveled by vehicles in the i-th traffic sub-segment after a traffic disturbance occurs. Let α represent the distance traveled by vehicles in the i-th traffic segment before the traffic disturbance occurs, cos represents the cosine function, and α represent the distance traveled by vehicles in the i-th traffic segment before the traffic disturbance occurs. i This represents the directional offset angle of the i-th traffic sub-segment.

6. The deep learning-based smart city traffic planning method according to claim 5, characterized in that: The specific steps in S2 also include: S23. Collect vehicle driving data under normal driving conditions in the activity area to calculate the path disturbance offset coefficient Py using sample data, and perform statistical analysis. Select the 90th percentile path disturbance offset coefficient Py as the path disturbance threshold Pyyz. For traffic paths within the activity area, compare and analyze the maximum path disturbance offset coefficient Pymax with the path disturbance offset coefficient Py to assess the traffic status of the activity area. The specific assessment content is as follows: If the maximum path disturbance offset coefficient Pymax ≥ path disturbance threshold Pyyz, then the traffic status of the activity area is determined to be abnormal, and traffic light cycle rhythm analysis is required. If the maximum path disturbance offset coefficient Pymax is less than the path disturbance threshold Pyyz, then the traffic status of the active area is determined to be normal, and no processing is required.

7. The deep learning-based smart city traffic planning method according to claim 6, characterized in that: The specific steps of S3 include: S31. When the traffic status in the activity area is abnormal, based on the light control cycle data, identify the midpoint between the green light end time Tend and the green light start time Tstart of each traffic light as the rhythm center point Tmid. S32. Define each pair of adjacent traffic lights as a traffic signal pair. For each traffic signal pair, calculate the difference in the rhythm center point ΔTmid between the two traffic lights within the same signal control cycle. Divide the difference in the rhythm center point ΔTmid by the response time difference Δτ to obtain the phase change rate of the traffic signal pair. Calculate the average phase change rate and construct the phase misalignment propagation index Xw. Here, the response time difference Δτ refers to the rhythm response time difference between traffic lights after a traffic disturbance event occurs. The rhythm response time refers to the time taken for a single traffic light to complete rhythm adjustment after a traffic disturbance event occurs, which is obtained through traffic signal control logs.

8. The deep learning-based smart city traffic planning method according to claim 7, characterized in that: The specific steps of S4 include: S41. Based on the disturbance area map Ga and the GIS network, the exit points and entrance points within the activity area are combined to construct OD pairs. The K-shortest path algorithm is then applied to each OD pair to identify the total number of traffic paths M within the activity area. Furthermore, the mean path disturbance offset coefficient Pyavg of the traffic paths within the activity area is calculated. The mean path disturbance offset coefficient Pyavg and the phase misalignment propagation index Xw are then dimensionlessly processed and coupled to obtain the comprehensive rhythm adaptation index Zh. The formula for calculating the comprehensive rhythm adaptation index Zh is as follows: In the formula, ln represents the natural logarithm function, Pyavg represents the mean path disturbance offset coefficient of traffic travel paths within the activity area, exp represents the exponential function, and Xw represents the phase misalignment propagation index. S42. Historical traffic light cycle data and vehicle driving-related data for the activity area are obtained from the traffic data storage center and used as sample data to calculate the comprehensive rhythm adaptation index Zh. The 10th percentile and 80th percentile of the comprehensive rhythm adaptation index Zh are selected as the first rhythm adaptation assessment threshold Zhyz1 and the second rhythm adaptation assessment threshold Zhyz2. The first rhythm adaptation assessment threshold Zhyz1 and the second rhythm adaptation assessment threshold Zhyz2 are compared and analyzed with the comprehensive rhythm adaptation index Zh to evaluate the operating status of the traffic signal control system and generate system adjustment commands. The specific evaluation content is as follows: If the comprehensive rhythm adaptation index Zh ≥ the second rhythm adaptation assessment threshold Zhyz2, then the traffic signal control system is determined to be in the third risk state. At this time, the traffic signal control system is in normal operation and a third system adjustment command is generated. If the first rhythm adaptation assessment threshold Zhyz1 < the comprehensive rhythm adaptation index Zh < the second rhythm adaptation assessment threshold Zhyz2, then the traffic signal control system is determined to be in the second risk state. At this time, the traffic signal control system has a local misalignment and generates a second system adjustment command. If the comprehensive rhythm adaptation index Zh ≤ the first rhythm adaptation assessment threshold Zhyz1, then the traffic signal control system is determined to be in the first risk state. At this time, the operation of the traffic signal control system is unbalanced, and the first system adjustment command is generated.

9. The deep learning-based smart city traffic planning method according to claim 8, characterized in that: The specific steps of S5 include: S51. Based on the operational status evaluation results of the traffic signal control system, intelligent adjustment of the traffic signal control system is performed using system adjustment commands, wherein the system adjustment commands include third system adjustment commands, second system adjustment commands, and first system adjustment commands. The third system adjustment command refers to maintaining the current operating status of the traffic signal control system and adjusting the data monitoring frequency of the traffic signal control system to 150% of the standard data monitoring frequency; The second system adjustment command refers to the adjustment of the signal control cycle of the misaligned signal pair. The signal control cycle adjustment refers to the preset phase change threshold. For traffic signal pairs in the activity area, the phase change rate is compared and analyzed with the phase change threshold. If the phase change rate is ≥ the phase change threshold, the traffic signal pair is identified as a misaligned signal pair. The green light time of the second traffic light in the misaligned signal pair is pushed back by 3 seconds, and the green light time window is extended by 2 seconds from the center time point to both sides. The first system adjustment command refers to signal control cycle reconstruction. Signal control cycle reconstruction involves constructing a propagation path on the traffic topology for identified misaligned signal pairs, forming a directional graph structure Gz, where Gz = (V, E), where V represents the intersection with the misaligned signal pair and E represents the adjacent intersection. Feature data for each traffic light node over the past N signal control cycles is collected, including signal control cycle duration, green light offset, average speed of vehicles passing through on green, and queue residual length at the end of green, to construct a feature vector F. Based on the graph structure Gz and the feature vector F, a graph neural network is used for inference training. A function model for rhythm optimization learning is constructed using traffic queue length, rhythm offset, and average vehicle travel delay as adaptive constraint parameters. This model outputs the optimal signal control cycle for each traffic light node, generating traffic light control commands for intelligent control of the traffic lights.

10. A deep learning-based smart city traffic planning system, applied to the deep learning-based smart city traffic planning method described in any one of claims 1-9, characterized in that: It includes a data acquisition module, a traffic status analysis module, a cycle rhythm analysis module, a comprehensive evaluation module, and an intelligent adjustment module; The data acquisition module is used to determine the activity area of ​​traffic disturbance activities through urban social media platforms, so as to construct a disturbance area map Ga, and to collect vehicle driving-related data and traffic light control cycle data within the disturbance area map Ga to construct a standardized dataset H; The traffic status analysis module is used to identify traffic travel paths based on the disturbance area map Ga, and analyze the changes in path length and direction deviation of vehicles before and after the start of traffic disturbance activities in order to assess the traffic status of the activity area, and to perform traffic light cycle rhythm analysis when the traffic status of the activity area is in an abnormal state. The cycle rhythm analysis module is used to calculate the difference in the rhythm center point of traffic lights based on the light control cycle data when the traffic status in the activity area is abnormal, and to quantify the rhythm consistency and response coordination between traffic lights. The comprehensive evaluation module is used to identify the total number of traffic travel paths M based on the disturbance area map Ga using the K-shortest path algorithm, analyze the operating status of the traffic signal control system within the activity area, classify the risk level, and generate system adjustment instructions for different risk levels. The intelligent adjustment module is used to intelligently adjust the traffic signal control system according to the system adjustment instructions.

Images