Smart city traffic optimization method and system based on 5G network

Abstract

The invention discloses a smart city traffic optimization method and system based on a 5G network, relates to the technical field of smart city traffic management, and is used for solving the problems of information consistency and conflict governance in city traffic flow regulation and control and shared parking space guidance. Comprising the steps of uploading of traffic flow and signal phase states, path guidance and berth recommendation in combination with berth occupancy states and vehicle owner position information, and berth state conflict recognition and partition induction treatment of a multi-source heterogeneous platform. And constructing a berth state behavior sequence, a heterogenous contention index, a state disturbance propagation graph and a conflict score cluster, and executing state shielding and path correction on high-conflict berths. Meanwhile, vehicle owner and vehicle behaviors are collected to complete parking use recording and settlement and illegal parking identification and processing, and a reservation authority management mechanism of the shared parking is constructed. The method is suitable for optimization induction under urban traffic dynamic management and control and parking space resource multi-platform access.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart city traffic management, more specifically, the present application relates to a smart city traffic optimization method and system based on 5G network. BACKGROUND

[0002] Under the current background of smart city construction, urban traffic optimization has gradually evolved from static strategy to real-time dynamic scheduling. The existing system generally collects traffic flow and traffic signal state by laying sensors, and combines with the navigation platform to recommend the path, but there are mainly three problems: First, the collection cycle, transmission delay and data consistency of traffic flow and signal state are different, causing the prediction model and real-time state to be mismatched, affecting the accuracy of signal timing strategy. Second, the multi-source heterogeneous access of shared parking platform has not established a unified information index structure, and each platform lacks unified standards in state marking, reporting interval and permission adjustment rules, resulting in problems such as permission mismatch and state drift in the process of parking state induction and pushing. Third, the responsibility of parking conflict induction is not clear, and the root cause of the induction failure cannot be analyzed and the strategy traced, and there is a lack of reconstruction mechanism for multi-platform behavior model.

[0003] Some current solutions attempt to use historical state and prediction model to establish congestion prediction mechanism, or introduce parking heat index to optimize recommendation results, but fail to deeply analyze the formation mechanism of induction conflict, especially lack of systematic analysis and behavior pattern extraction of multi-platform state label switching lag, and cannot perform fine-grained state shielding and path correction for high contention areas, thereby affecting the scheduling efficiency of parking resources and user experience.

[0004] In view of the above problems, the present application provides a solution. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide a smart city traffic optimization method and system based on 5G network to solve the problems proposed in the background art.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solutions: In a preferred embodiment, comprising: Collecting traffic flow and signal phase state data, uploading and caching through 5G network, and establishing a data set with device number, time stamp and geographic location index; Cleaning, feature extraction and modeling of uploaded traffic data, generating traffic prediction results and traffic state labeling, identifying congestion intersections and performing signal timing adjustment; Collect the parking space occupation state and establish a parking space state pool, combine the vehicle owner's location with the parking space attributes to calculate the path, generate parking space recommendations and push them to the terminal, and publish the guidance information at the same time; Build a parking space state behavior sequence and a heterogenous contention index, identify conflict potential areas and divide state contention sub-cluster, perform state shielding and path correction on high conflict parking spaces, carry out guidance deviation simulation and platform responsibility fitting, collect vehicle entry and exit records to complete settlement and illegal parking processing, and establish a shared parking reservation and permission management mechanism.

[0007] In a preferred embodiment, traffic flow sensors and traffic signal monitoring equipment are deployed to collect traffic flow and signal phase state data, which is uploaded at high speed through the 5G network, edge forwarded and classified cached, and a structured data set indexed by device number, timestamp and geographic location is established for format verification and abnormal isolation processing.

[0008] In a preferred embodiment, in the cloud data processing stage, traffic perception data is cleaned, feature extracted and classified modeled to generate traffic flow prediction results and traffic state labels for each path, and based on the predicted values, identify the intersections that will be congested and include them in the signal optimization object to improve traffic efficiency by adjusting the green light duration.

[0009] In a preferred embodiment, the occupancy state of each parking space is collected and reported, a parking space state pool is constructed and multi-dimensional indexing is established, and based on the current location of the vehicle owner and the parking space attributes, path calculation and accessibility determination are performed to generate parking space recommendation results and push them to the user terminal, while the remaining parking space information in the area is published through the guidance screen.

[0010] In a preferred embodiment, in the cloud parking space state pool, search for historical events related to failed guidance, compare the guidance generation time and platform permission response time, extract permission state mismatch samples, classify and analyze the behavior characteristics of each platform in state label switching, reporting interval and authorization adjustment period, construct a drift coupling curve reflecting the response lag law, and based on the multi-platform state fluctuation overlap interval, derive the minimum observation time window.

[0011] In a preferred embodiment, within the minimum observation time window, the state upload behavior sequence of each parking space is constructed, a multi-dimensional index is extracted to generate a heterogenous contention index, and the temporal coupling relationship between parking spaces is analyzed to generate a state disturbance propagation graph and a guidance conflict basis subgraph, and through spatial mapping and multi-source index superposition, identify conflict guidance potential subareas, and finally cluster and divide parking spaces according to conflict scores to form multiple state contention sub-clusters.

[0012] In a preferred embodiment, on the basis of berth conflict score and state contention sub-cluster division, a heterogenous contention index threshold is set, state shielding or path penalty correction is implemented on the berth with contention index exceeding the threshold and in a high conflict sub-cluster, and participation weight of the berth in navigation path planning is adjusted; In each conflict-induced potential partition, an induced task deviation cluster, a platform rewriting hypothesis and a hypothesis drift path set are constructed, parallel simulation analysis is carried out, the path model with the minimum deviation is selected by comparing the induced deviations under each hypothesis path, and the responsibility proportion of the master platform is fitted in reverse according to the state change information and the induced response record of the optimal drift path, and the state uploading and permission synchronization strategy adjustment suggestions for the master platform are generated.

[0013] In a preferred embodiment, the vehicle owner and vehicle information are bound, the vehicle access time is recorded to generate a berth use record and complete settlement, the illegal parking behavior is identified and a reminder is pushed, and if there is no response, an illegal parking record is generated and enters law enforcement scheduling, and a shared berth platform is constructed to support reservation and use permission management.

[0014] In a preferred embodiment, it comprises: a vehicle flow acquisition and data structuring module, a cloud analysis and signal scheduling module, a berth state recommendation and induction publishing module, and a berth conflict identification and comprehensive management module, and the modules are signal connected; The vehicle flow acquisition and data structuring module is mainly used for acquiring vehicle flow and signal phase state data, uploading and caching through a 5G network, and establishing a data set with device number, time stamp and geographic location index; The cloud analysis and signal scheduling module is mainly used for cleaning, feature extraction and modeling of uploaded traffic data, generating vehicle flow prediction results and traffic state labels, identifying congestion intersections and executing signal timing adjustment; The berth state recommendation and induction publishing module is mainly used for acquiring berth occupancy state and establishing a berth state pool, performing path calculation combined with vehicle owner position and berth attribute, generating berth recommendations and pushing to the terminal, and publishing induction information; The berth conflict identification and comprehensive management module is mainly used for constructing berth state behavior sequence and heterogenous contention index, identifying conflict potential areas and dividing state contention sub-clusters, executing state shielding and path correction on high conflict berths, carrying out induced deviation simulation and platform responsibility fitting, collecting vehicle owner access records to complete settlement and illegal parking processing, and establishing a shared berth reservation and permission management mechanism.

[0015] The technical effects and advantages of the intelligent city traffic optimization method and system based on a 5G network are as follows: The application realizes unified caching and format checking of high-frequency perception data by constructing a structured index system of traffic flow and signal state data; in combination with a cloud prediction modeling mechanism, early identification and timing strategy scheduling are performed on congested intersections. In terms of parking space guidance, a parking space state pool and path accessibility judgment mechanism are established, and precise recommendation and guidance information are published in combination with user positions. On this basis, an induction failure behavior sample identification mechanism is proposed, and a drift coupling curve and a minimum observation time window are constructed based on the state switching lag feature, effectively restoring the platform response delay mode; relying on the state upload behavior sequence, a heterogeneous contention index is generated, the conflict structure between parking spaces and the guidance propagation path are analyzed, and the parking spaces are clustered and divided according to the conflict score, and state shielding and path weight adjustment are implemented in the conflict partition, realizing dynamic guidance control of parking space resources. Finally, the parking space usage record generation and settlement linkage are completed in combination with the vehicle owner behavior data, and a shared parking space reservation and permission management mechanism is constructed, completing the whole cycle optimization control. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A timing diagram of the intelligent city traffic optimization method and system based on the 5G network.

[0017] Figure 2 A module schematic diagram of the intelligent city traffic optimization method and system based on the 5G network. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application. EMBODIMENT

[0019] The application discloses an intelligent city traffic optimization method based on a 5G network, as shown in the figure, comprising the following steps: Figure 1 Collecting traffic flow and signal phase state data, uploading and caching through a 5G network, and establishing a data set with device number, time stamp and geographic location index; Cleaning, feature extraction and modeling the uploaded traffic data, generating traffic prediction results and traffic state labels, identifying congested intersections and performing signal timing adjustment; Collecting parking space occupation states and establishing a parking space state pool, calculating paths in combination with vehicle owner positions and parking space attributes, generating parking space recommendations and pushing them to terminals, and publishing guidance information at the same time; ​The state behavior sequence of the parking space and the heterogeneity contention index are constructed, the conflict potential area is identified, and the state contention sub-cluster is divided. The state shielding and path correction are performed on the high conflict parking space. The deviation simulation and platform responsibility fitting are carried out. The entry and exit records of the vehicle owner are collected to complete the settlement and illegal parking processing. The shared parking reservation and permission management mechanism is established.

[0020] In dynamic traffic operation management, first of all, the continuous perception of the road traffic state is realized, and the global traffic monitoring task is completed relying on the new generation of low delay and high rate 5G communication means. In this process, through the deployment of networked perception devices and the construction of data channels with edge access capability, high-frequency perception and real-time reporting of the running state of key traffic nodes are realized.

[0021] Specifically, first of all, for traffic hubs, road intersections and main road sections, traffic flow sensors and traffic signal monitoring devices are deployed according to the principle of covering the main traffic paths. Among them, the traffic flow sensor adopts a combination scheme of microwave induction device and laser radar with traffic detection special protocol to collect the number of vehicles passing through the section per unit time; the traffic signal monitoring device obtains the current signal phase state by accessing the output port of the traffic controller, and is accompanied by the time sequence identification within the signal light period. All networked perception devices are set by their internal firmware control program to set the acquisition frequency and timestamp recording rules. The acquisition frequency is matched according to the road grade and traffic pressure level, and the timestamp information is generated by a local high-precision clock chip and written into each data record, ensuring that the time reference is unified and traceable.

[0022] Further, in order to realize high-speed transmission and low delay upload, each networked perception device accesses the nearest 5G base station node through its integrated 5G communication terminal in a wireless manner. The access action follows the low latency uplink scheduling mechanism defined in the 5G NR protocol cluster, and completes the device identification binding and data structure packaging at the physical channel layer. The data collected by the device is automatically encoded into a data frame with a timestamp, while retaining its original structure and accompanying source device ID and location identification, and is uploaded in real time through the 5G channel.

[0023] After uploading, the 5G base station pushes the data to the high-performance edge server deployed at the edge location through the preset edge forwarding rules.

[0024] In this processing node, first of all, according to the data type field carried in the data frame, automatic classification is performed, and the traffic volume data, signal phase data and time sequence data are respectively stored in different logical buffer queues.

[0025] Subsequently, the data partition structure initializer is called with the data type as the index dimension, the sub-datasets for different purposes are constructed, and the multi-field index table structure is generated in each sub-dataset, and the index fields include: device number, collection timestamp, geographic location information and device category mark.

[0026] It should be noted that the above data set construction process adopts hash mapping combined with B+ tree index structure to realize high-frequency access path optimization, which ensures that the subsequent modeling algorithm can retrieve the required historical time series records in milliseconds. At the same time, in order to ensure the consistency of data structure, format verification logic is executed during the writing of each data, which checks the field integrity and time stamp continuity in real time, and triggers an exception mark and peels off abnormal data when missing fields or time dislocation are found, and writes to an asynchronous isolation buffer to prevent it from affecting the integrity of the formal data set.

[0027] In the cloud data processing stage, the data cleaning program is first called for pre-processing operation on the perception data. This program takes single-path traffic data as input, sets the unique timestamp index as the discrimination criterion, and sequentially eliminates duplicate records, peels off abnormal samples and fills in missing value gaps; The completion logic of missing data is based on the combination of time series moving average and nearest neighbor interpolation, and immediately performs integrity check after completion to ensure the continuity of the basic data for subsequent statistical analysis.

[0028] The cleaned data stream is grouped according to the path number, and the sliding window of the past n sampling points is extracted in each group to calculate the average traffic, peak traffic and minimum traffic in the corresponding time period. The above three statistical indicators are used as the basic characteristic quantity of each path.

[0029] Next, the above statistical characteristics and the signal phase state corresponding to the current period are combined to construct a multi-dimensional input vector.

[0030] In this combination structure, the signal phase state represents the current state and duration of the red, green and yellow lights at each intersection in the form of encoding, and all vectors are arranged in time step order to form a training sample sequence. And through standardization processing, each characteristic quantity is mapped to a unified numerical interval to ensure that the model input features have balanced numerical distribution.

[0031] Subsequently, the LSTM modeling channel is called to construct the training set with the input sequence, and the model is iteratively trained according to the time-reversed error gradient, and the Adam optimizer is used to control the convergence rate during training. After the model training is completed, the future traffic flow prediction value corresponding to one time is directly output, and the intersection number and timestamp are recorded in the prediction result buffer.

[0032] To further identify the typical traffic patterns of different road operating periods, the clustering channel is called to classify the feature results of the modeling output. Specifically, the K-means clustering algorithm is used, and the average traffic flow and signal duration of each path in a certain period are used as the clustering dimensions to construct the feature vector input.

[0033] Meanwhile, in the clustering operation, the cluster center is initialized and multiple iterations are performed to minimize the Euclidean distance as the objective function for sample classification until the cluster center stabilizes. Then the final output of the cluster label is defined as three state types of traffic peak, valley and normal, and the label result is mapped back to the original path-time index to form a state labeling dataset, providing conditional branch basis for matching traffic control strategies in different regions and different time periods.

[0034] Further, after completing the operating state recognition, the predicted traffic flow is used as the input source, and the preset traffic flow first threshold is used for comparison and judgment. If the predicted value exceeds the threshold and the corresponding path is in the marked interval of the cluster label for traffic peak state, the intersection is automatically marked as “soon to be congested” state, and its path number is written into the optimization target queue as the strategy action point for subsequent signal adjustment.

[0035] For the identified congestion trend section, based on the premise that the total duration of red, green and yellow lights in the current period remains unchanged, the green light duration allocation ratio is dynamically adjusted. Specifically, more green light time is allocated to the passing direction with higher LSTM prediction value, so as to improve the effective traffic flow per unit time.

[0036] The upper limit constraint of green light duration is introduced in the adjustment process to prevent green signal over-allocation from causing congestion spread to adjacent intersections, while ensuring that the safety interval between signal phases meets the standard requirements of the minimum delay between adjacent phases in the traffic signal control specification.

[0037] In the static traffic resource allocation, first, at the entrances and exits of each parking lot and roadside public parking spaces, geomagnetic sensors, infrared beam sensors and video stream analysis terminals are preferentially deployed.

[0038] Among them, the geomagnetic sensor detects whether a vehicle has entered the parking space by detecting the disturbance change of the geomagnetic field; the infrared beam device detects whether the infrared beam is blocked to determine the vehicle entry and exit state; the video stream analysis terminal calls the embedded image recognition algorithm to form continuous video stream input at the deployment location, extracts the vehicle body contour and edge features to determine whether the parking space is occupied. All sensing devices are pre-integrated with 5G communication terminals and automatically access the 5G access point in the corresponding area through the set wireless configuration parameters to realize millisecond-level uploading of vehicle entry and exit events.

[0039] The raw data uploaded by the perception device includes event type, timestamp, parking space number, and physical location coordinate information, and is written into the cloud parking space state pool in real time after being transmitted through the 5G channel.

[0040] In the parking space state pool, a multi-dimensional data index is constructed according to the parking space number and geographic location information. The data index structure establishes a joint primary key through the parking space ID, administrative region code, and GPS position fields, and adds a state update time and data source marker field to each parking space state record to ensure timeliness and traceability.

[0041] Subsequently, to realize the parking space navigation recommendation function, the city navigation engine is called and the current location data of the vehicle owner is synchronously obtained. The vehicle owner's location data is obtained through mobile terminal authorization, and the collection method relies on the GPS coordinates output by the user terminal positioning function. The navigation engine takes the current location of the vehicle as the starting point of the path, and takes all parking spaces marked as "idle" in the parking space state pool as candidate endpoints. The path planning model calculates the path time consumption of the vehicle from the current location to each candidate parking space. At the same time, the path planning model is based on the A* path algorithm, and combines real-time traffic data and no-entry rules to output the shortest expected travel time of each path, and sorts the paths according to the travel time.

[0042] Further, after the candidate path sorting is completed, the parking space attribute analysis logic is called to perform reachability verification on each candidate parking space. The reachability verification logic includes whether the parking space is in a restricted area, whether it requires special access permissions, and whether the current user has access permissions.

[0043] After verification, the parking space charging strategy analysis function is further called to analyze the charging rules of the parking space in the current period and to calculate the joint score of path time consumption, reachability result, and expected cost. Finally, the parking space with the highest comprehensive score is selected as the recommended target.

[0044] Among them, the special access permissions include intra-unit parking spaces and community closed parking spaces; the charging rules include time-of-use charging, period cap, and first-hour free.

[0045] Subsequently, after the target parking space is calculated and confirmed, the parking space number, path guidance data, and parking space state details are pushed to the user terminal through the navigation API interface. If the user terminal is a car navigation system, the navigation path is received through the CAN bus interface and displayed through a graphical interface. If it is a mobile terminal device, the path is displayed and the target parking space information is marked through a mobile APP calling a map SDK, supporting real-time viewing and confirmation by the user.

[0046] Meanwhile, in order to assist the users of vehicles that have not entered the navigation state to obtain the parking space supply information in the region, electronic parking guidance screens are arranged along the main traffic trunk roads in the city. The data required by the guidance screens is derived from the remaining parking space data in the parking space state pool. The system divides the region according to the administrative division or functional block, and the number of parking spaces marked as free in each region is summarized, and a "region number-remaining parking space number" data structure is formed, which is refreshed to the guidance screen end through the information publishing program in a timely manner.

[0047] It should be noted that in the scenario of multi-agent accessing shared parking space resources, there are parking space state information reported by multiple source systems from government agencies, commercial properties, community management platforms and the like, and there is no unified standard for data collection frequency, permission state marking, reservation strategy and state return rule of each parking space management party, especially in the dimensions of data upload interval, state synchronization trigger mechanism and authorized strategy adjustment period, etc. In the prior art, the parking space state is collected by various sensing devices, uploaded to the cloud state pool through the 5G network, and indexed in multiple dimensions according to the parking space number and geographical position information for navigation path planning calling, but there is no unified analysis and integration for data consistency, state drift history and management permission difference, resulting in that in the guidance screen pushing or navigation recommendation, the same parking space is displayed as available in the navigation engine, but actually cannot be used due to the platform not reporting the permission change in time before the user arrives; the navigation path planning is guided to the parking space region where the state abnormally occurs, causing high-frequency detour and parking space competition of vehicles; and some high-permission parking spaces are missing in the guidance system for a long time, causing reduced resource utilization. Therefore, in the embodiment, firstly, in the cloud parking space state pool corresponding to the 5G network, all historical events in which a conflict occurs between the actual parking position of the guidance path and the final availability of the parking space in the past continuous T sampling periods are searched.

[0048] The conflict event set is generated according to the parking space state update record and the parking position coordinate matching logic of the guidance recommendation record, and is composed of a ternary index structure of parking space number, timestamp and platform identifier, which is used to locate the data source and response process corresponding to each guidance task.

[0049] After obtaining the complete conflict event set, the platform state label, guidance publishing platform identifier, permission check result and platform definition of the parking space associated with each event in the guidance generation are stripped out. The guidance publishing platform identifier and the platform state label are derived from the guidance task generation log, and the log field includes platform ID, guidance publishing time, guidance target parking space number and platform record state before publishing; the permission check result is implemented by extracting the fields in the platform response result log structure, and the specific fields include permission request action response code, parking space reservation mark and call return state value. All log data has complete time sequence identifier and is recorded to a unified time reference in a synchronous manner, so as to facilitate subsequent error measurement.

[0050] Further, after obtaining all the conflict sample data, by comparing the induction generation time point with the timestamp of the permission state response record, all sample sets caused by the failure of the induction task due to the failure of the master platform to upload permission changes in time are identified, and a permission state mismatch sample set is formed. For each event in the permission state mismatch sample set, further locate the time offset between the induction task generation time point and the corresponding permission state latest update time point, and construct a permission adjustment delay mapping table to reflect the specific offset behavior of platform response lag.

[0051] Subsequently, the above time offset samples are classified and grouped according to the platform to which they belong, and the following three indicators are extracted for each platform respectively: The state label switching amplitude per unit time is defined as the ratio of the number of continuous change events to the time length; The state reporting interval sequence dispersion is defined as the standard deviation of the upload interval time within T periods; The authorization policy adjustment period is defined as the minimum cycle length of policy changes in the platform permission control logic.

[0052] Next, the three indicators are respectively assigned stability, responsiveness and structural weight, and a weighted integral expression is constructed to complete the platform drift coupling curve modeling under the exponential weighted moving average rule with time as the horizontal axis and the induction response mismatch probability as the vertical axis.

[0053] The calculation expression of the platform drift coupling curve in this embodiment is: Dp(t)=Σ{k=0}^{T}λ×(1-λ)^k×δp(t-k); Wherein, Dp(t) represents the state response drift intensity of platform p at time t; δp(t-k) is a binary indicator of whether there is a permission state mismatch at time t-k (1 for mismatch, 0 for consistency); λ is the exponential decay factor, which is system default configuration or calibrated by historical samples.

[0054] Wherein, each platform drift coupling curve is used to express the probability evolution trajectory of the platform in any historical time period due to the lag of permission response leading to induction response mismatch. After modeling is completed, the induction task start time point corresponding to the first induction failure event in the platform drift coupling curve is extracted, and the state record sequence of the platform is traced back to mark the most intensive continuous interval of state fluctuations before that time point. At the same time, in all platforms, cross compare the most intensive state fluctuation sections of this type, extract the minimum time segment length overlapping between them, and define it as the minimum observation time window.

[0055] wherein the derivation formula of the minimum observation time window Tmin is as follows: Tmin=max{Δti|i∈[1,N]}, wherein Δti represents the maximum change interval of traffic flow per unit time in the ith observation period in the same lane.

[0056] By sampling all Δti sequences in historical data and selecting the maximum value as the lower limit of the window, it is ensured that the model input can cover at least one complete change period, avoiding the problems of data sparsity or pattern truncation in model training.

[0057] It should be noted that since the generation of the minimum observation time window must rely on the dynamic linkage behavior of the permission control boundary not unified among multiple platforms, the state label granularity different, the upload protocol response rules different and the synchronous participation in the induction trigger mechanism, it only has a defined meaning under the structural condition that multiple heterogeneous platforms participate in berth induction navigation but there is no unified state constraint rule. Once it is out of this type of induction system structure, it will not be able to complete the generation of the multi-platform drift coupling curve and the extraction of the offset closed interval, so the time scale itself loses definability.

[0058] Next, the minimum observation time window length is defined as N1, and a sliding time window with a time length of N1 is set as the unified time domain reference for the analysis of the historical state behavior of each berth. On this basis, all state upload records within the past N1 time length are retrieved in units of berths, and the upload behaviors are constructed in chronological order to form a behavior sequence matrix, which is used to depict the dynamic response trajectory of the berth within the window and the behavior characteristics of the heterogeneous platforms.

[0059] In the behavior sequence matrix, each upload behavior is composed of five tuples, including the source platform identifier, the upload frequency, the platform state label at the time of induction generation, the permission change marker and the call response result.

[0060] Specifically, the source platform identifier is used to indicate which berth management platform reported the data, the upload frequency is obtained by counting the number of uploads per unit time, the state label is extracted from the state field of the calling platform at the time in the induction task log, the permission change marker is calculated by comparing the change trend of the permission state field in the continuous records, and the call response result is read from the response field of the induction publishing platform at the time of request initiation. The above five fields all have a timestamp index structure and are filled into the matrix in order.

[0061] After the behavior sequence matrix is constructed, the heterogeneous contention index of each berth is further calculated to represent the competition degree and conflict potential between the upload behaviors of each platform within the N1 window. Specifically, the heterogeneous contention index is calculated by fusing the following three quantitative factors: State stability factor, defined as the reciprocal of the number of state label switching times within the N1 time window, reflecting the stability of the parking space state in this period; Master platform lag delay coefficient, defined as the matching rate of platform authority state and user reservation behavior, that is, whether the state provided by the platform is consistent with the user's operable authority between the formation of the induction call initiation and the reservation response, as a lag indicator of response credibility; Upload behavior drift tensor, defined as the shortest path distance of state transition graph structure submitted by different source platforms, to measure the time difference between platform behaviors.

[0062] Wherein, the calculation expression of the heterogenous contention index is: Ci=α×(1 / (nsi+1))+β×(1−mri)+γ×dij; nsi represents the number of state label switching times of the parking space within the N1 window, that is, the state stability factor; mri represents the permission matching rate of the parking space, defined as the ratio of the number of successful matches to the total number of inductions, that is, the master platform lag delay coefficient; dij represents the shortest drift path distance of platform upload behavior in the state transition graph structure, that is, the upload behavior drift tensor; α represents the state stability factor weight coefficient, β represents the master platform lag delay coefficient weight coefficient, and γ represents the upload behavior drift tensor weight coefficient, which is set by system default configuration or according to empirical data; Subsequently, with N1 sliding time window as the anchor point, the state label change trajectory, upload behavior trigger time sequence, permission control label and induction call path of all parking spaces in this period are uniformly called and synchronized on the time axis, and the response coordination analysis basis across parking spaces is constructed.

[0063] On this basis, the coupling relationship between each parking space and its surrounding parking spaces in the following three dimensions is calculated respectively: The consistency of state label switching direction; The synchronization of switching frequency; The coincidence of upload response delay time; And based on the above three types of coupling relationship, a four-dimensional interaction tensor structure is constructed, which is a four-dimensional interaction tensor field with parking space number, upload trigger time, state switching event identifier and permission change label as four dimensions, to capture the heterogenous time coupling characteristics of induction response behavior within the N1 window between parking spaces.

[0064] Subsequently, taking the berth as the node, the state disturbance propagation graph generation algorithm is called, the edge weight value is generated according to the coupling degree between each dimension in the above-mentioned tensor, and the dynamic connection is established between the berth nodes to form a directed weighted graph expressing the time sequence coupling relationship of the heterogeneous behaviors. And on the structure of the directed weighted graph, the minimum weight edge adsorption strategy is executed, the node aggregation is performed on the berth set with the same main control platform and the response coupling degree in the N1 window being significantly higher than the average level of the neighborhood, a group of conflict-induced basic subgraphs are generated, which are used as the initial structure basis for subsequent spatial analysis and conflict partition.

[0065] Then, referring to these conflict-induced basic subgraphs, spatial mapping is performed in the map coordinate domain, and the spatial distribution model of the graph nodes is generated according to the actual geographical position projection of the berth. On this basis, the upload abnormal frequency distribution graph in the 5G network and the navigation induction path error density heat map are superimposed, and the superimposition is realized through unified grid partition and berth number-grid mapping relationship. Finally, multiple regions with high state coupling density and high induction error frequency characteristics are identified, which are defined as conflict induction potential partitions.

[0066] In each conflict induction potential partition, further taking the berth as the basic unit, the state abnormal trigger times, the number of induction failures and the heterogeneous contention index value of the berth in the N1 time window are called, wherein the three indexes respectively reflect the state fluctuation activity, the task failure sensitivity and the platform conflict density.

[0067] Subsequently, a multi-factor linear fusion function is called to weight and synthesize the three indexes, generate the berth conflict score of each berth in the potential partition, and perform clustering division on all berths by taking the score as the input to form multiple state contention sub-class clusters. The berth conflict score is calculated in the embodiment as follows: Si=w1×ei+w2×fi+w3×ci; Wherein, ei is the state abnormal trigger times of the berth in the N1 time window; fi is the number of induction failures, that is, the number of samples that the actual berth cannot use after induction; ci is the heterogeneous contention index; w1, w2, w3 are weight coefficients set by the scheduling strategy, used to express the relative importance of each factor in the conflict intensity.

[0068] The clustering process adopts the K-means algorithm, and the input features are the conflict score Si of each berth and the spatial density index of the berth in the conflict induction potential partition. The initial cluster number is determined by the elbow method ElbowMethod, and the k value corresponding to the inflection point is selected according to the change trend of the cluster number k and the total squared error SSE. The clustering results serve as the structural basis for navigation path planning and state shielding, ensuring that conflict berths are effectively classified and managed.

[0069] Next, based on the completion of berth conflict score and state contention sub-cluster division, in order to avoid the phenomenon of concentrated landing points of navigation path in the area with high incidence of incorrect guidance, and further to cause the navigation engine to form incorrect guidance in the state uncertain area, a state shield for the induction engine is further constructed, taking the alien contention index as the core trigger condition, and performing state intervention and path correction operations on the berth sub-cluster where the conflict behavior frequently occurs.

[0070] Specifically, first, the alien contention index threshold is set, wherein the alien contention index threshold is derived from the mean and variance of the distribution of the berth contention index of all samples in the N1 sliding window, and the alien contention index threshold interval is set by setting multiple standard deviation levels.

[0071] When the contention index of a berth exceeds the set alien contention index threshold and is in the identified state contention sub-cluster, the state shielding strategy is triggered, and the current state of the berth in the berth state pool is marked as a shield state and is automatically excluded in navigation path planning. If the berth does not meet the complete shielding condition but is still in the conflict strong correlation area, the path penalty weight of the berth is calculated by taking the berth conflict score as the input, and the cost function of the berth in the path planning model is dynamically improved, so that the navigation path actively avoids such berth nodes under the condition of similar shortest path, thereby completing the path penalty correction.

[0072] Further, in each induction conflict potential partition, three parallel simulation flows are periodically executed: state response backtracking, induction result deviation simulation, and master platform drift simulation. At the same time, the parallel simulation flow generates a multi-path parallel analysis link by taking the historical induction task execution record as the input benchmark.

[0073] In the state response backtracking phase, first, all samples in which the induction landing position and the final berth availability are inconsistent are retrieved from all executed induction tasks, and an induction result deviation cluster is constructed, wherein the induction result deviation cluster is a four-tuple data structure consisting of berth number, induction execution time, user path ID, and final state label, serving as a benchmark deviation source.

[0074] On this basis, a master platform rewriting hypothesis set is constructed. Specifically, by simulating the scenario of changes in the master control right of the induction task publishing right of each platform, multiple state rewriting models are generated. In each model, the induction task is controlled by different platforms, and their respective state update mechanisms are re-implanted to form alternative state paths. In the state flow, the rewriting paths are embedded in the original induction state chain according to the original time sequence and state logic, generating a set of hypothesis drift paths.

[0075] Subsequently, within a parallel simulation environment, the simulation engine was invoked to simulate the induced effects of all the aforementioned hypothetical drift path sets. The simulation process used the induced execution path as the primary index and the final berth availability tag as the outcome variable. By comparing the induced deviation rates under each drift path, the path model with the minimum deviation was statistically analyzed. A weighted similarity analysis algorithm was then used to construct a multidimensional similarity matrix between the historical state label sequence, the permission synchronization time offset sequence, and the user reservation execution results. Ultimately, the drift path with the best similarity was selected as the optimal master control logic simulation model for the potential partition under the current state structure.

[0076] Finally, based on the three indicators of the platform authority update time point corresponding to the optimal drift path, the state label reaction lag degree and the induced response efficiency, the reverse responsibility fitting operation is performed on the current main control platform.

[0077] The specific method is to construct a platform responsibility function, with the platform response lag as the input variable and the induced deviation probability as the response function output, to fit the contribution ratio of platform status lag to navigation engine failure, and generate the main control platform status update strategy recommendations based on this, including: upload cycle adjustment recommendations, permission synchronization window expansion recommendations and induced task approval logic postponement recommendations, etc., which are output in text form for the management to receive and execute.

[0078] The platform responsibility fitting function is defined as follows in this embodiment: Rp=μ1×τp+μ2×εp+μ3×θp; in, τp is the average platform state response delay; εp is the platform induction deviation rate, that is, the proportion of samples with failed platform control induction; θp is the percentage of platform permission synchronization timeouts; μ1, μ2, and μ3 are weight coefficients in the responsibility sharing model.

[0079] In the parking fee collection process, in order to realize the automated processing of parking space usage records and financial settlement, car owners are first required to complete the registration operation through the front-end interface before using the platform service, and then call the registration process interface to collect the car owner's vehicle information and payment account information, and perform identity authentication and account binding verification. After registration is completed, a unique owner ID and vehicle ID binding comparison table is generated and stored in the user data pool.

[0080] When a vehicle enters or exits a parking lot, a video recognition device preset at the entrance or exit captures the image stream at a fixed frame rate, executes the license plate recognition algorithm, extracts the license plate number, and calls the owner comparison table to complete identity confirmation.

[0081] The entry time is bound to the license plate number and recorded as the starting time marker of the parking space usage. After the vehicle leaves, the video recognition device completes exit recognition again and records the end time. The time difference between the two records is the parking space usage duration. The parking space usage duration data is input into the billing calculator, and the current parking space billing policy template is used to calculate the payment. The payment is automatically deducted from the bound payment account and the transaction status information is returned.

[0082] Finally, the parking space usage record and payment data are synchronized and written into the cloud data archiving area according to the timestamp index. The structure field includes the owner ID, parking space number, entry and exit time, usage duration, charging policy version number, and settlement result status. This realizes the unified archiving and calling of parking space resource usage information and financial turnover information.

[0083] In terms of order maintenance, the parking video recognition device deployed on both sides of the road continuously collects parking image streams along the line and executes illegal parking detection algorithms. The illegal parking detection algorithm combines the parking area boundary model and the parking authorization map to determine the spatial overlap between the spatial position of the vehicle appearing in the video and the current legal parking space library. If the detection result is an illegal occupation area, mark the illegal parking event in the image recognition result. At the same time, write the illegal parking event into the dispatch buffer queue, immediately call the owner registration information, and push the warning reminder to the owner through the short message API and voice reminder channel.

[0084] Further, the idle parking space library interface is called to filter legal available parking spaces within the current area range, and a quick recommendation algorithm based on path distance and parking space score is executed to push the optimal parking path and state to the owner through the mobile terminal interface, prompting the owner to leave the illegal area as soon as possible and go to the recommended parking space.

[0085] The reminder process sets the default response period to 8 minutes. After sending the reminder, start the countdown monitoring logic. If no vehicle departure action is detected within the specified period, automatically enter the illegal parking record processing flow and generate an illegal parking record form to write into the law enforcement pending queue. The back-end dispatch logic calls the law enforcement task scheduling table to mark the corresponding road section as a high-priority patrol section, and the back-end law enforcement personnel handle the illegal parking vehicles offline combined with the patrol track.

[0086] For vehicle owners who have violated parking multiple times or have bad faith, a bad faith factor is constructed by accumulating the number of illegal parking times, the number of illegal parking non-response times, and the historical response period timeout ratio, and a bad faith vehicle behavior record feedback mechanism is formed. For users who reach the bad faith threshold, their use of some high-level public parking space resources is restricted through platform permission control logic, realizing the governance intervention of credit constraints in the static resource configuration process.

[0087] Furthermore, to improve the utilization efficiency of parking resources, especially the sharing rate of vacant parking spaces at night, a unified shared parking platform has been established. This platform allows government agencies and community property management to set sharing time periods, access permissions, and reservation status through a management interface. The platform updates the shared availability mark based on the parking space number and makes it available through the user app for query and reservation.

[0088] After a user completes a reservation, the reservation information is automatically recorded and a reminder is pushed before the shared time period. The frequency of shared berth usage and the reservation ratio are simultaneously counted. Incentive rules are used to award points or subsidies to property owners to promote sharing, forming a positive closed loop of resource openness and profit feedback.

[0089] The present invention also proposes a smart city traffic optimization system based on 5G network, such as Figure 2 As shown, it includes: traffic flow collection and data structuring module, cloud analysis and signal scheduling module, berth status recommendation and induction release module, and berth conflict identification and comprehensive management module, and the signal connections between each module.

[0090] The vehicle flow collection and data structuring module is mainly used to collect vehicle flow and signal phase status data, upload and cache it through the 5G network, and create a data set with device number, timestamp and geographic location index; The cloud-based analysis and signal scheduling module is mainly used to clean, extract features and model uploaded traffic data, generate traffic flow prediction results and traffic status annotations, identify congested intersections and perform signal timing adjustments; The berth status recommendation and guidance release module is mainly used to collect berth occupancy status and establish a berth status pool. It combines the vehicle owner's location and berth attributes to calculate the path, generate berth recommendations, and push them to the terminal, while also publishing guidance information. The berth conflict identification and comprehensive management module is mainly used to construct berth state behavior sequences and heterogeneous contention indexes, identify conflict potential areas and divide state contention subclass clusters, perform state shielding and path correction on high-conflict berths, conduct induced deviation simulation and platform responsibility fitting, collect vehicle owner entry and exit records to complete settlement and illegal parking processing, and establish a shared berth reservation and authority management mechanism.

[0091] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0092] The above embodiments may be implemented in whole or in part through software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0093] Those skilled in the art can understand that the modules and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software manner depends on the specific application of the technical solution and the constraints of the invention. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0094] In addition, each functional module in each embodiment of the present application can be integrated in one processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0095] The above is merely specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0096] Finally: the above is only the preferred embodiment of the present application, and is not used to limit the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A smart city traffic optimization method based on 5G network, It is characterized by including: Collect vehicle flow and signal phase status data, upload and cache it via the 5G network, and create a data set with device number, timestamp, and geographic location index; Clean, extract features, and model uploaded traffic data to generate traffic flow predictions and traffic status annotations, identify congested intersections, and adjust signal timings. Collect parking space occupancy status and establish a parking space status pool. Combine the vehicle owner's location and parking space attributes to calculate the route, generate parking space recommendations, push them to the terminal, and publish guidance information. Construct a parking space status behavior sequence and heterogeneous contention index, identify conflict potential areas and divide status contention sub-clusters, perform status shielding and path correction on high-conflict parking spaces, conduct induced deviation simulation and platform responsibility fitting, collect vehicle owner entry and exit records to complete settlement and illegal parking processing, and establish a shared parking space reservation and authority management mechanism.

2. The method for optimizing smart city traffic based on a 5G network according to claim 1, characterized in that: Deploy traffic flow sensors and traffic signal monitoring equipment to collect traffic volume and signal phase status, and upload, edge forward, and cache the data at high speed through the 5G network. At the same time, establish a structured data set indexed by device number, timestamp, and geographic location for format verification and exception isolation.

3. The method for optimizing smart city traffic based on 5G network according to claim 1, characterized in that: During the cloud-based data processing phase, traffic perception data is cleaned, features extracted, and classified into models to generate traffic flow prediction results and traffic status annotations for each path. Based on the predicted values, intersections that are about to be congested are identified and included in the signal optimization objects, thereby improving traffic efficiency by adjusting the green light duration.

4. The method for optimizing smart city traffic based on 5G network according to claim 1, characterized in that: Collect and report the occupancy status of each parking space, build a parking space status pool and establish a multi-dimensional index, perform path calculation and accessibility judgment based on the owner's current location and parking space attributes, generate parking space recommendation results and push them to the user terminal, and publish the remaining parking space information in the area through the guidance screen.

5. The method for optimizing smart city traffic based on 5G network according to claim 1, characterized in that: Historical events related to induced failure are retrieved from the cloud berth status pool. By comparing the induction generation time with the platform permission response time, permission status mismatch samples are extracted. The behavioral characteristics of each platform in terms of status label switching, reporting interval and authorization adjustment cycle are classified and analyzed. A drift coupling curve reflecting the response lag law is constructed, and the minimum observation time window is derived based on the overlapping interval of multi-platform status fluctuations.

6. The method for optimizing smart city traffic based on 5G network according to claim 5, characterized in that: Within the minimum observation time window, the state upload behavior sequence of each berth is constructed, and multi-dimensional indicators are extracted to generate a heterogeneous contention index. Then, the temporal coupling relationship between berths is analyzed, and a state disturbance propagation diagram and an induced conflict basic subgraph are generated. The conflict-inducing potential partitions are identified through spatial mapping and superposition of multi-source indicators. Finally, the berths are clustered according to the conflict scores to form multiple state contention subclass clusters.

7. The method for optimizing smart city traffic based on 5G network according to claim 6, characterized in that: Based on the berth conflict score and state contention sub-cluster division, a heterogeneous contention index threshold is set. For berths whose contention index exceeds the threshold and are in high-conflict sub-cluster, state blocking or path penalty correction is implemented, and their participation weight in navigation path planning is adjusted. In each conflict-induced potential partition, we construct induced task deviation clusters, platform rewriting hypotheses, and hypothetical drift path sets, and carry out parallel simulation analysis. By comparing the induced deviations under each hypothetical path, we select the path model with the minimum deviation. Based on the state change information and induced response records of the optimal drift path, we reversely fit the responsibility ratio of the master control platform and generate adjustment suggestions for the state upload and authority synchronization strategy for the master control platform.

8. The method for optimizing smart city traffic based on 5G network according to claim 1, characterized in that: Collect information about the owner and the vehicle to establish a binding, record the vehicle's entry and exit time to generate a parking space usage record and complete settlement, identify illegal parking behaviors and push reminders, if there is no response, generate an illegal parking record and enter law enforcement scheduling, build a shared parking platform to support reservations and usage rights management.

9. A smart city traffic optimization system based on 5G network, characterized by: include: Traffic flow collection and data structuring module, cloud analysis and signal scheduling module, berth status recommendation and guidance release module, and berth conflict identification and comprehensive management module, with signal connections between each module; The vehicle flow collection and data structuring module is mainly used to collect vehicle flow and signal phase status data, upload and cache it through the 5G network, and create a data set with device number, timestamp and geographic location index; The cloud-based analysis and signal scheduling module is mainly used to clean, extract features and model uploaded traffic data, generate traffic flow prediction results and traffic status annotations, identify congested intersections and perform signal timing adjustments; The berth status recommendation and guidance release module is mainly used to collect berth occupancy status and establish a berth status pool. It combines the vehicle owner's location and berth attributes to calculate the path, generate berth recommendations, and push them to the terminal, while also publishing guidance information. The berth conflict identification and comprehensive management module is mainly used to construct berth state behavior sequences and heterogeneous contention indexes, identify conflict potential areas and divide state contention subclass clusters, perform state shielding and path correction on high-conflict berths, conduct induced deviation simulation and platform responsibility fitting, collect vehicle owner entry and exit records to complete settlement and illegal parking processing, and establish a shared berth reservation and authority management mechanism.

Images