A dynamic profiling and trusted updating system for heterogeneous traffic data

Abstract

The application discloses a kind of dynamic filing and trusted updating systems of heterogeneous traffic data, it is related to intelligent transportation technology field, specifically includes: definition traffic management and traffic video private network, with license plate, identity, enterprise characteristic code as three-dimensional association key, combine hash matching, track similarity algorithm and multi-factor dynamic weighting strategy Fusion multi-network data, generate real-time holographic archives, extract road network five-dimensional features, through decision tree matching core scene model set, generate early warning information through scene priority weighting algorithm, and with disposal feedback dynamic optimization model parameter, based on reinforcement learning and monte carlo tree search optimization scheduling path and resource allocation, in combination with Kalman filter pre-judgment target trajectory, linkage traffic video private network monitoring time sequence tracking and signal dynamic green wave timing, according to risk level Generation differentiation strategy, realize research and judge, scheduling, control seamless connection, improve the risk research and judge accuracy and scheduling disposal efficiency, reduce road network influence.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation technology, specifically relating to a dynamic archiving and reliable updating system for heterogeneous traffic data. Background Technology

[0002] In current traffic information processing, structured data from different sources exhibit significant differences in terms of time validity, field completeness, source reliability, and the tightness of object correlation. Existing archive construction methods typically employ static aggregation or single-time data entry mechanisms, lacking continuous data quality assessment and tiered processing mechanisms. This results in low-quality data directly participating in archive updates, easily leading to distorted archive content, unstable correlations, and significant fluctuations in subsequent analysis results.

[0003] Furthermore, existing technologies often resort to direct discarding or manual review when faced with data whose relationships cannot be confirmed temporarily. This results in low automation and makes it difficult to balance the real-time nature and accuracy of data updates. Especially in scenarios with continuous access to multi-source data, effective solutions remain lacking for controlling the risk of erroneous writes while ensuring update efficiency.

[0004] Existing risk assessment models are mostly general-purpose designs, failing to build dedicated model sets for core traffic management scenarios such as key vehicle control, accident hazard warning, congestion prediction, and chain reactions of illegal events. They also fail to dynamically match model parameters with real-time road network characteristics such as regional type, traffic flow, time period, weather, and construction. As a result, the models have poor adaptability and lack a scientific integration and priority determination mechanism for the output results of multiple models, making it difficult to guarantee the confidence of the assessment results. Furthermore, a closed loop for model optimization based on actual handling feedback has not been established, and the model parameters and decision thresholds are fixed, making it impossible to continuously improve the assessment accuracy and resulting in a high rate of false alarms and false negatives.

[0005] Therefore, there is an urgent need for a dynamic archiving and reliable update system for heterogeneous traffic data to achieve quality assessment of access data, hierarchical management of candidate data, and dynamic correction of archive content, thereby improving the accuracy and stability of archive construction. Summary of the Invention

[0006] The purpose of this invention is to provide a dynamic archiving and reliable update system for heterogeneous traffic data, which solves the technical problems of insufficient security, efficiency and collaboration in cross-domain data and command interaction, weak multi-domain heterogeneous data fusion and governance capabilities, and lack of real-time holographic archive support in the existing technology.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A dynamic archiving and reliable updating system for heterogeneous traffic data, characterized in that it includes: The traffic data interaction gateway defines the traffic management network and the traffic video network. It pushes structured perception data from the traffic video network to the traffic management network through an encrypted one-way import channel, and issues control commands to the execution devices of the traffic video network through a trusted reverse security proxy channel after authorization verification and auditing. The dynamic holographic archive construction module connects to the traffic data interaction gateway, integrates multi-domain data, and completes data governance and generates real-time updated holographic archives through three-dimensional correlation and multi-factor dynamic weighting strategies. The multi-source risk assessment module matches model parameters according to real-time road network characteristics, integrates the outputs of multiple models such as time series prediction and spatial clustering, and generates early warning information containing risk level, cause and related files through a scenario priority weighting algorithm. It uses the handling feedback as a supervision signal to dynamically update feature weights and decision thresholds, and continuously improves the accuracy of assessment. The command and dispatch module presents the risk situation by overlaying early warning information, archive data and cross-network resource status. With the goal of optimal handling efficiency and minimal impact on the road network, it dynamically optimizes dispatch paths and resource allocation by combining real-time traffic flow and resource status, generates differentiated strategies, automatically triggers traffic video network monitoring and tracking, green wave timing of traffic signals, and synchronizes with traffic management network instructions to achieve seamless connection between analysis, dispatch and control.

[0008] Furthermore, the traffic management private network and the traffic video private network are defined, and the specific methods are as follows: The traffic management private network refers to a dedicated network domain used by traffic management departments to carry core business systems and process structured business data. The traffic video private network refers to a dedicated network domain built independently of the traffic management private network, used to access front-end video image acquisition equipment, carry real-time video stream data, and perform video image analysis.

[0009] Furthermore, the three-dimensional association in the dynamic holographic archive construction module specifically includes: Using license plate number, identity feature code, and enterprise feature code as association keys, a hash matching algorithm is used to quickly compare the license plate number captured by the traffic video network with the vehicle registration data of the traffic management network. If the match is successful, the identity feature code and enterprise feature code associated with the license plate number are extracted from the business database to form a temporary three-dimensional association relationship. If the match fails, the license plate number to be matched and its continuous spatiotemporal trajectory are stored in the waiting queue, and the similarity is calculated with the de-identified vehicle trajectory features in the historical archive, and the association relationship is dynamically updated.

[0010] Furthermore, the multi-factor dynamic weighting strategy in the dynamic holographic archive construction module is specifically as follows: Set data weight evaluation factors, assign basic credibility weights to data from different sources, set data freshness weights that decay exponentially with the interval between the data generation time and the current time, calculate data integrity weights in reverse based on the proportion of missing fields, and assign data relevance weights based on the degree of association with the three-dimensional association key. The comprehensive weight of each data point is calculated using a linear weighted summation formula. Data with a comprehensive weight higher than a preset threshold are directly used for file updates, while data with a weight lower than the threshold are stored in a verification queue. The final decision on whether to include the data in the file is made after verification with subsequent related data.

[0011] Furthermore, the model parameters are matched according to real-time road network characteristics, specifically as follows: The system pre-defines four core scenario model sets: key vehicle control, accident hazard warning, congestion prediction, and chain reaction of illegal events. Each model set contains multiple parameter templates adapted to different road network characteristics. It extracts road network feature parameters in real time, including area type, traffic flow level, time period attributes, weather conditions, and road construction status. The system uses a decision tree algorithm to classify real-time road network feature parameters and match them with the corresponding scenario model set. Based on the historical accuracy and real-time data freshness in that scenario, the system dynamically adjusts the model parameters through an adaptive weighted algorithm. The weights for historical accuracy and data freshness are reverse-corrected based on the assessment results in the most recent period, thus achieving dynamic optimization of the parameter templates.

[0012] Furthermore, warning information is generated using a scene priority weighted algorithm. The specific method is as follows: A basic priority is assigned to each scenario model. The confidence level of each model's output is calculated based on the model's historical accuracy, input data completeness, and feature matching degree. The comprehensive priority of each model's output is calculated using a weighted formula of basic priority × confidence level. For conflicting judgment results, the result with the highest comprehensive priority is taken as the core warning conclusion. The core conclusion is combined with supplementary information from other model outputs to generate complete warning information.

[0013] Furthermore, the scheduling path and resource allocation are dynamically optimized, specifically through the following methods: A scheduling decision model is constructed using reinforcement learning algorithms. Real-time traffic flow, police location and load, monitoring point coverage, and traffic signal control authority are used as the state space. A reward function is designed with response time, road network traffic efficiency loss, and police resource utilization rate as the core reward factors. The optimal scheduling path and resource combination scheme are explored through genetic algorithms to generate differentiated strategies.

[0014] Furthermore, the automatic triggering of traffic video network monitoring and tracking, and green wave timing of traffic signals, is achieved through the following methods: Based on the location and movement trajectory of the risk target in the early warning information, the Kalman filter algorithm is used to predict the target's route and the time of arrival at each intersection. Traffic video network monitoring points along the route are selected, and monitoring and tracking instructions are triggered in sequence according to the target's arrival time. At the same time, the monitoring images are transmitted back in real time. According to the predicted arrival time and real-time traffic flow, the start time and duration of the green light at each intersection are calculated using the green wave timing algorithm to generate a dynamic green wave scheme. The green wave scheme and the police dispatch route are issued synchronously and pushed to the traffic video network traffic signal controller through the trusted reverse security proxy channel of the traffic data interaction gateway. The timing parameters are dynamically fine-tuned according to the actual speed of the target to ensure smooth handling channels.

[0015] Furthermore, traffic video surveillance points along the route are selected, and monitoring and tracking instructions are triggered in sequence according to the target's arrival time. The specific method is as follows: Based on the predicted trajectory of the target and the spatial topology of the road network, monitoring points covering the route and without obstructions are selected. Priority is set according to point resolution and monitoring angle, and sorted based on the estimated arrival time of the target. A time-series pre-triggering algorithm is employed, with pre-set trigger thresholds to achieve seamless tracking of points. If the target deviates from the trajectory, points are re-selected in real time based on the latest location, and the triggering order is dynamically adjusted. In summary, due to the adoption of the above technical solution, the beneficial effects of this invention are: 1. This invention evaluates multi-source structured data from multiple dimensions, including source reliability, time validity, field integrity, and object correlation, thereby achieving data quality screening before archive updates, reducing the probability of low-quality data being directly written into archives, and improving the accuracy and stability of archive content. 2. By setting up a queue to be verified, this invention retains and performs secondary verification on data that cannot be fully confirmed temporarily but has potential value, thus avoiding the information loss caused by directly discarding usable data in the prior art. It ensures real-time performance while also taking into account the accuracy of updates. 3. By utilizing feedback from historical verification results, this invention can dynamically adjust the weight allocation of each evaluation factor, enabling the archive update mechanism to have adaptive optimization capabilities and improving the long-term availability of the system in continuous access scenarios. This invention does not depend on specific business processes or single application scenarios and can serve as a data support module for upper-level early warning analysis, behavior recognition, trend analysis, and other applications, exhibiting good versatility and scalability. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This invention illustrates a module diagram of a dynamic archiving and reliable updating system for heterogeneous traffic data. Figure 2 The flowchart of the present invention for dynamic archiving and reliable updating of heterogeneous traffic data is shown. Figure 3 A flowchart of the multi-domain data fusion and archive update method of the present invention is shown. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] like Figure 1 , Figure 2 The system shown is a dynamic archiving and reliable updating system for heterogeneous traffic data, which specifically includes the following: The traffic data interaction gateway defines the traffic management network and the traffic video network. It pushes structured perception data from the traffic video network to the traffic management network through an encrypted one-way import channel, and issues control commands to the execution devices of the traffic video network through a trusted reverse security proxy channel after authorization verification and auditing. The traffic management network refers to a dedicated network domain used by traffic management departments to carry core business systems and process structured business data. This includes systems such as the public security traffic management integrated application platform and the traffic integrated command platform, responsible for storing and processing business data such as vehicle registration, driver information, and traffic violations / accidents. The traffic video network refers to a dedicated network domain built independently of the traffic management network, used to access front-end video image acquisition equipment. This includes checkpoints, electronic police systems, high-point surveillance, traffic incident detection equipment, and their corresponding streaming media servers, video storage and analysis servers. It carries real-time video stream data and performs structured analysis of video images, generating structured perception data such as vehicle passage records, violation captures, and incident detection results.

[0020] The traffic data exchange gateway includes an encrypted one-way import channel and a trusted reverse security proxy channel. The encrypted one-way import channel securely pushes structured perception data from the traffic video private network to the traffic management private network. Specifically, the data source within the traffic video private network encrypts the structured perception data packets using the national cryptographic SM4 algorithm, with a 128-bit encryption key. Before encryption, a cyclic redundancy check (CRC) code is appended to the data packet. The CRC32 code is calculated using the formula CRC32 = Hash(data content + timestamp), where the timestamp is the UTC time at the time of data generation and is used for integrity verification at the receiving end. The encrypted data packet is transmitted to the data receiving server of the traffic management private network via a physical one-way import device. This device ensures that data can only flow from the traffic video private network to the traffic management private network and cannot be transmitted in the reverse direction. The receiving end performs CRC32 verification on the data packet. If the verification passes, it decrypts the data using the same key and stores the data in a buffer for subsequent processing. If the verification fails, it initiates a retransmission request to the traffic video private network data source through a trusted reverse security proxy channel. The retransmission interval increases in increments of 1 second, 3 seconds, and 5 seconds, with a maximum retransmission limit of 3 times to ensure the integrity and reliability of data transmission.

[0021] The trusted reverse security proxy channel is used to send control commands generated by the traffic management private network to the execution devices of the traffic video private network after authorization verification and auditing. In practice, a three-level authorization mechanism of "role-permission-device" is used to assign differentiated control permissions to different users. For example, the command center administrator can issue signal control commands, while the brigade's on-duty personnel can only access video surveillance resources. When a user issues a command, the system attaches a digital signature to the command content. The signature is calculated using the national cryptographic SM2 algorithm, with the formula: Sign = SM2(Command Content + Role Identifier + Timestamp). The command content consists of executable control commands issued from the traffic management network to the traffic video network, including monitoring control commands (such as starting monitoring and tracking, switching screens), signal control commands (such as green wave timing parameters, phase adjustment), etc. The role identifier represents a unique string identifier assigned by the system to different operating roles (such as ADM-TRAFFIC-001 for the command center administrator and DUTY-TEAM-012 for the brigade duty personnel), which is bound to the role's permissions. The timestamp represents the UTC timestamp when the command is generated and issued on the traffic management network. The digital signature (Sign) is a string generated by encrypting the combination of "command + role identifier + timestamp" using the SM2 asymmetric encryption algorithm. It is used by the receiving end to verify the legality of the command's source and the integrity of its content. After the command is verified by the signature and matched with the permissions, it is directed to the traffic signal controllers and monitoring equipment on the traffic video network, ensuring the non-repudiation and integrity of the command.

[0022] When instructions are transmitted to the traffic video private network via a reverse security proxy channel, the receiving end first verifies the validity of the digital signature, and then queries the authorization database based on the role identifier to confirm whether the user has the operation permission to operate the target execution device. After the permission verification is successful, the instruction is pushed to the corresponding execution device, such as the traffic signal controller, monitoring pan-tilt unit, or variable message sign. The device executes the corresponding action after receiving the instruction. All cross-network operations generate audit logs, which include the operation role, operation time, instruction content, target device identifier, etc. The logs are retained for one year and support querying and tracing by time, role, device, and other dimensions, enabling rapid location and auditing of abnormal operations.

[0023] The dynamic holographic archive construction module connects to the traffic data interaction gateway, integrates multi-domain data, and completes data governance and generates real-time updated holographic archives through three-dimensional correlation and multi-factor dynamic weighting strategies. like Figure 3 As shown, it integrates three types of core data: business data such as vehicle registration, driver information, and violations and accidents from the traffic management network; structured perception data such as vehicle passage records, violation captures, and event detection results pushed through an encrypted one-way import channel from the traffic video network; and supplementary data such as GPS data of key vehicles and waybills from the external network. This achieves comprehensive integration of data from multiple network domains and provides a complete data source for subsequent governance. Association key definition: The three-dimensional association key is the vehicle's unique identifier (such as the de-identified license plate hash value), the driver's unique identifier (such as the identity feature code), and the organization's unique identifier (such as the enterprise feature code); A hash matching algorithm is adopted. The algorithm performs Hash(identifier) ​​= MD5(identifier + dynamic salt value) on each type of association key data. The dynamic salt value is generated daily (calculated from the system timestamp and device code hash). Here, the identifier specifically refers to the specific data item corresponding to the above three-dimensional association key, namely vehicle unique identifier data (such as desensitized license plates captured by the traffic video network, vehicle registration number of the traffic management network), driver unique identifier data (such as identity feature code of the traffic management network, associated desensitized identity identifier), and unit unique identifier data (such as enterprise feature code, unit code of key vehicles). The hash matching algorithm calculates the hash value of the identifier to be matched (such as vehicle unique identifier) ​​and compares it with the corresponding identifier hash value stored in the historical archive (hash value of the identifier to be matched / hash value of the corresponding identifier stored in the historical archive). The comparison result is expressed as a similarity percentage. In this embodiment, the matching threshold is set to 98%. When the percentage is <98%, it is judged as a matching failure, that is, the basic association relationship between people, vehicles and enterprises cannot be directly established. When a match fails, the association is completed using a trajectory spatiotemporal similarity algorithm. The algorithm formula is as follows: ,in This represents the trajectory vector of the vehicle to be matched in the traffic video network (in this embodiment, the dimension = number of trajectory points × 2, and each trajectory point is composed of latitude and longitude coordinates). The trajectory vector (dimension and dimension) of the target vehicle in the historical archive. Consistent) The spatial weight coefficients are determined using an adaptive optimization algorithm. This indicates the duration of the temporal intersection between the trajectory to be matched and the historical trajectory. This represents the length of the maximum temporal intersection in the historical trajectory database.

[0024] Furthermore, regarding the spatial weighting coefficient In this embodiment, the initial value is set to 0.6, and the objective function is F = number of hits / total number of matches, with the trajectory matching accuracy over the past 30 days as the objective function. The total number of matches refers to the total number of association matches performed using the spatiotemporal similarity algorithm over the past 30 days (including trajectory comparison attempts for all vehicles to be matched). The number of hits refers to the number of times that the matching attempts were confirmed as correctly associated through subsequent cross-network data verification (such as consistency between traffic video network capture features and traffic management network registration information) or feedback from frontline handling. The value is iteratively adjusted using the gradient descent method. Where t represents the iteration round (t increments from 0, with each iteration corresponding to 1 iteration). (Parameter adjustment) The learning rate is used to calculate the objective function F after each iteration. In this embodiment, it is set as follows: The convergence is reduced to the 0.7±0.03 range to ensure that spatial trajectory matching has a higher priority than temporal overlap.

[0025] The credibility weight W1 of traffic management network business data is set by comprehensively considering multi-source features. For example, in this embodiment, W1 is set using a two-stage method of factor level mapping and dynamic constraint calibration. Multi-source features, such as data source authority, update frequency, and historical error rate, are each divided into three levels (Authority: Official core library = Level 3, Departmental business library = Level 2, Temporary cache library = Level 1; Update frequency: Real-time update = Level 3, Daily update = Level 2, Weekly update = Level 1; Historical error rate: ≤0.5% = Level 3, 0.5%-2% = Level 2, >2% = Level 1). A three-dimensional level-interval mapping matrix is ​​constructed and generated through training on historical credibility verification data. For example, (level 3, level 3, level 3) corresponds to [0.42, 0.45], (level 3, level 3, level 2) corresponds to [0.40, 0.43], (level 2, level 2, level 2) corresponds to [0.37, 0.40], and (level 1, level 1, level 1) corresponds to [0.35, 0.38]. The basic credibility interval is directly matched by the combination of factor levels. The hit rate of the past 30 days of data is introduced as a constraint factor. If the hit rate is ≥90%, the upper limit of the basic interval is increased by 0.01; if the hit rate is <70%, the lower limit of the basic interval is decreased by 0.01. Finally, W1 is the median of the calibrated interval to ensure that the weights both conform to the characteristics of the factors and are suitable for the actual application effect.

[0026] A scenario-based exponential decay model is used to set the freshness weight W2 of the traffic video network sensing data, and the formula is as follows: ,in Generate a time interval between the current and current data. The attenuation coefficient is used for reverse calibration based on the hit attenuation pattern in risk assessment using historical data. Based on the impact of data integrity on the decision-making accuracy of the two scenarios, the scenario suitability coefficient was determined after scenario-based needs surveys and verification of historical decision-making effects. Such as risk assessment scenarios =1.2, Command and Dispatch Scenario =0.9, using the formula The data integrity weight is represented by Q, where Q represents the number of core fields and QZ represents the number of missing fields, which is determined by filtering through frequently accessed fields. First, the association level is divided according to the degree of matching of the three-dimensional association key (direct association = 1.0, indirect association = 0.5, no association = 0). Then, the relevance score between the data and the core business (obtained from historical data statistics) is combined. Using the formula W4 = association level × association level weight + relevance score × relevance score weight, the relevance weight of the external private network data is set as W4. The association level weight and the relevance score weight are determined by analyzing the data application effect in multiple scenarios and combining business adaptation verification, based on the need for a balanced contribution of the degree of matching of the association key and the relevance of the core business to the relevance assessment, to ensure that the two assessment dimensions have an equal impact on the weight result. The data value contribution is calculated based on the core module call hit data over the past 30 days. The calculation is based on the statistical period of the past 30 days, and the specific method of obtaining the data is as follows: Total number of calls: The system automatically records the number of times the core modules (multi-source risk assessment module and command and dispatch module) actively query this data, including all explicit access behaviors such as extracting data features during risk assessment and calling related information during command and dispatch, and accumulates the total number of calls; Hit count: Determined by the reverse effect of the core module's decision-making. If the risk assessment module generates a warning information based on the data after it is called and the feedback after handling is "hit", or if the command and dispatch module optimizes the solution based on the data and reduces the handling response time by ≥20% / road network efficiency loss by ≥15%, then it is counted as 1 hit, and the hit count is accumulated. = Number of hits / Total number of calls, with the overall weight calculated using W1, W2, W3, and W4. Set the overall weight threshold w. Data that is not verified is directly updated in the archive; otherwise, it is stored in the queue to be verified. The multi-source risk assessment module matches model parameters according to real-time road network characteristics, integrates the outputs of multiple models such as time series prediction and spatial clustering, and generates early warning information containing risk level, cause and related files through a scenario priority weighting algorithm. It uses the handling feedback as a supervision signal to dynamically update feature weights and decision thresholds, and continuously improves the accuracy of assessment. Five core road network features are extracted through multi-domain data fusion, including the regional type F1 from the road network basic database of the traffic management network (including road level and administrative division), the traffic flow level F2 from the traffic video network vehicle data (average traffic flow of road segments within 5 minutes) and traffic management network traffic statistics, the time period attribute F3 based on the system's local UTC time and the time period division rules of the traffic management network (peak / off-peak / night), the weather condition F4 from the external dedicated network meteorological data interface (real-time weather type and visibility), and the construction status F5 from the construction reporting database of the traffic management network and the event detection results of the traffic video network (construction area identification). The moving average method was used to denoise the traffic flow data, and the median imputation method was used to handle missing values ​​in the meteorological data and construction status data. All feature data were converted into a standardized format, and the time was uniformly aligned to the UTC timestamp. Labels were used to encode the regional type (F1: urban area=3, expressway=2, suburban area=1), time period attributes (F3: peak=3, off-peak=2, night=1), weather conditions (F4: sunny=1, cloudy=2, rain=3, snow=4, fog=5), and construction status (F5: no construction=1, construction=2). Traffic flow level F2 is discretized according to the design capacity of road segments (for example, setting smooth traffic = 1: traffic flow < 60% of design capacity; slow traffic = 2: 60%-85% of design capacity; congested traffic = 3: > 85% of design capacity). The final output is a quantized feature vector [F1, F2, F3, F4, F5]. The feature extraction frequency is consistent with the data acquisition cycle (real-time extraction, updated every 30 seconds).

[0027] The design scenario model set includes four core scenario model sets, each with five sets of dynamic parameter templates. The key vehicle control model is adapted for highways, urban areas, and school perimeters, with core warning threshold parameters of 0.8 for highways, 0.7 for urban areas, and 0.65 for school perimeters. The accident hazard warning model is adapted for intersections, road sections, and construction areas, with core spatiotemporal clustering radius parameters of 50m for intersections and 100m for road sections. The congestion prediction model is adapted for peak, off-peak, and holiday scenarios, with core LSTM network neuron count parameters of 256 for peak hours and 128 for off-peak hours. The illegal event chain reaction model is adapted for main roads, commercial districts, and urban-rural fringe areas, with core chain reaction radius parameters of 800m for main roads and 500m for commercial districts.

[0028] The quantized feature vectors are classified, and the decision tree splitting criterion is as follows: ,in This represents the characteristic attribute of the current split (i.e., one of F1-F5). The feature sample set of the current node. Information gain refers to the feature attribute. For feature sample set The reduction in information entropy of the sample set after partitioning is used to measure the classification ability of the feature attributes. Representing splitting information, referring to characteristic attributes For feature sample set The breadth and uniformity of the partitioning are used to correct for bias in information gain. This represents the gain ratio, used to address the issue that information gain tends to select feature attributes with a larger number of values.

[0029] Based on historical accuracy A (number of hits in the last 30 days / total number of warnings) and real-time data freshness F (normalized value of the time interval Δt between data generation time and the current time), parameters are adjusted through an adaptive weighted algorithm, as shown in the formula: ,in The initial template parameters are calibrated based on historical data training results from similar smart transportation projects, combined with the road network characteristics and traffic management needs of the area to be managed. It is a weighting coefficient for the accuracy of historical judgments. It is a weighting coefficient for the freshness of real-time data, determined based on correlation analysis between multi-scenario experimental data and road network assessment results. This is a dynamic correction term, determined through reverse calibration based on historical model performance indicators such as false alarm rate and hit count. These are the application parameters ultimately used in the model.

[0030] First, basic priorities are assigned according to the urgency of the scenario. Then, confidence is calculated by combining the model's historical accuracy, the completeness of the input data, and the feature matching degree. A comprehensive priority is obtained by weighting the basic priority and the confidence and incorporating a risk superposition correction term. For multi-model conflict warnings, the result with the highest comprehensive priority is taken as the core conclusion. After integrating supplementary information, a complete warning information containing risk level, cause, etc. is generated. Furthermore, the basic priority allocation in this embodiment is as follows: accident hazard early warning P0=0.4, key vehicle control P0=0.3, congestion prediction P0=0.2, and chain reaction of illegal events P0=0.1, using the formula The confidence score is calculated, where A is the historical judgment accuracy, C is the input data completeness (C = 1 - number of missing fields / total number of fields), where the total number of fields is the total number of fields corresponding to the entity in the dynamic holographic archive, M is the cosine similarity between the target feature and the model training feature, and c1, c2, and c3 are the corresponding weights. These weights are determined based on the core impact of the model's historical accuracy on the reliability of the judgment, the fundamental supporting role of the input data completeness, and the auxiliary verification value of the feature matching degree, combined with industry practice and multiple rounds of model training and verification.

[0031] Using formula Calculate the overall priority, where As a risk overlay correction item, when there are multiple risk overlays (such as key vehicles + accident-prone road sections + peak hours), set RiskAdjust=0.2; when there are no overlays, RiskAdjust=0.

[0032] For multi-model conflict warnings for the same target (such as simultaneously triggering key vehicle control and violation event warnings), the result with the highest comprehensive priority P is taken as the core conclusion. Supplementary information from other models (such as violation type, road section risk level, and vehicle historical violation records) is integrated to generate complete warning information. In this embodiment, the risk level is divided according to the value of P. For example, the risk level is divided into high risk if P ≥ 0.7, medium risk if 0.5 ≤ P < 0.7, and low risk if P < 0.5.

[0033] The command and dispatch module presents the risk situation by overlaying early warning information, archive data and cross-network resource status. With the goal of optimal handling efficiency and minimal impact on the road network, it dynamically optimizes dispatch paths and resource allocation by combining real-time traffic flow and resource status, generates differentiated strategies, automatically triggers traffic video network monitoring and tracking, green wave timing of traffic signals, and synchronizes with traffic management network instructions to achieve seamless connection between analysis, dispatch and control.

[0034] Construct a scheduling decision model, optimize scheduling paths and resource allocation through Monte Carlo tree search, output differentiated strategies according to high, medium and low risk levels, and link cross-network monitoring point timing tracking and dynamic green wave timing of signal machines; Construct a scheduling decision model, where the state space S is defined as a high-dimensional vector. Where T is the real-time traffic flow (including average vehicle speed and traffic volume of the road segment). This refers to the status of police forces (including location, load, and equipment). To monitor the coverage area of ​​the monitoring points, For signal control authority. The level of road network congestion; Action space B includes three types of actions: police force dispatch (including police force selection and route planning), monitoring call (including location selection and tracking triggering), and signal control (including green wave timing and phase adjustment). Reward function formula ,in To handle response time, To improve the utilization rate of police resources, This results in a loss of road network traffic efficiency. To improve cross-network resource collaboration efficiency, These are the corresponding weights, and the weight values ​​are determined based on the priority of core traffic management needs, ensuring that the sum of the weights is 1. The response time is defined as the total time from when the system generates an early warning and issues a dispatch instruction to when the police arrive at the risk site and complete the preparation for handling. The data source is the GPS positioning data and operation timestamp of the police's mobile terminal, and the instruction issuance timestamp of the command and dispatch system. Police resource utilization rate is defined as the ratio of the actual working hours of the police officers involved in the response to the available working hours of that police officer on that day. The data source is the scheduling data and response work records of the police force management system. The road network traffic efficiency loss is defined as the percentage decrease in the average vehicle speed of road sections within the risk-affected area compared to normal times after the implementation of the scheduling plan. The data sources are vehicle passage data from the traffic video network and traffic flow monitoring data from the traffic management network. Cross-network resource coordination efficiency is defined as the ratio of the average delay of cross-network resources (monitoring, traffic signals) in responding to scheduling commands to a preset standard delay. The data source is the command transmission log of the traffic data interaction gateway. In this embodiment, the Monte Carlo tree search optimization involves initializing 100 populations (feasible scheduling schemes), selecting using roulette wheel selection, performing single-point crossover (crossover probability 0.7), and performing random mutation (mutation probability 0.05), iterating 100 times to explore the optimal scheduling scheme, and outputting differentiated strategies: for high-risk warnings, the police dispatch range is ≤3 kilometers, adopting a full-point monitoring linkage along the route, with the highest signal timing priority (exclusive green wave); for medium-risk warnings, the police dispatch range is ≤5 kilometers, adopting a key-point monitoring linkage, with medium signal timing priority (while also considering traffic); for low-risk warnings, a police patrol mode is adopted, triggering only core monitoring points, with low signal timing priority (routine fine-tuning).

[0035] The formula for predicting the target's route and arrival time based on the Kalman filter algorithm is as follows: ,in This represents the optimal predicted position of the target at time k. This indicates the predicted position at time k based on the data at time k−1. This represents the actual observed position of the target at time k (based on coordinate data collected from monitoring points on the traffic video network). The value is the observation matrix (taken as the identity matrix, since the position observation is a direct mapping of two-dimensional coordinates). Kalman gain; Filter monitoring points that cover the route and have no obstructions, according to a priority formula. Sort, where The resolution is set to 1.0 for 1080P and 0.7 for 720P in this embodiment. To ensure the monitoring angle adaptability (in this embodiment, 1.0 is set for the route directly facing the camera and 0.6 is set for the route to the side), These are the corresponding weights, and the actual weight values ​​are determined by linear regression fitting using measured data (target recognition success rate at different resolutions and angles). After sorting by estimated arrival time, a time-series pre-triggering algorithm is used, with an advance triggering time of [time value missing]. Where D is the actual distance between adjacent monitoring points (the straight-line distance is calculated from the coordinates of the points in the road network database), and V is the estimated speed of the target (calculated based on the instantaneous speed moving average of the vehicle passing data from the traffic video network, with a sliding window of 3 collection cycles, each cycle being 1 second). The fixed additional time represents the warm-up and parameter adjustment time for the monitoring equipment, which is determined by measuring the start-up response time of different types of monitoring equipment (electronic alarm, PTZ camera); When the target deviates from the trajectory, the similarity is recalculated based on the ratio of the current trajectory length to the predicted trajectory length, based on the latest position. The similarity threshold is determined based on the statistical analysis of historical trajectory deviation data. When the similarity is less than the corresponding similarity threshold, a re-filtering is triggered; otherwise, it is not triggered. The current trajectory length is the cumulative distance traveled by the target in real time, and the predicted trajectory length is the total distance of the initial predicted route. Both are obtained by summing the Euclidean distances calculated from the coordinate point sequence.

[0036] The dynamic optimization of green wave timing is based on the predicted arrival time of the target and real-time traffic flow. First, the arrival time of the target at the intersection is calculated by combining the distance from the target's current location to the intersection and the estimated speed. Then, the green light start time is determined based on this time. The lead time of the start time is dynamically adjusted according to the traffic flow in the target's direction of travel; the smaller the traffic flow, the smaller the lead time, and the larger the traffic flow, the larger the lead time (in this embodiment, the lead time ranges from 3 to 5 seconds). The green light duration is calculated by combining the traffic flow in the target direction, the queue length at the intersection, the average vehicle speed of the road segment, and the number of lanes to ensure that traffic demand is met. The system collects the actual travel speed of the target every 30 seconds and adjusts the green light duration accordingly. The adjusted duration is controlled between 0.8 and 1.2 times the original duration to avoid sudden timing changes affecting road network stability. The entire optimization process is constrained by a green wave bandwidth of no less than 30% to ensure green light coverage throughout the target's journey and guarantee unobstructed access.

[0037] The above formulas are all dimensionless calculations, and the preset parameters in the formulas should be set by those skilled in the art according to the actual situation.

[0038] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0039] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementations. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A dynamic archiving and reliable updating system for heterogeneous traffic data, characterized in that, include: The traffic data interaction gateway defines the traffic management private network and the traffic video private network, enabling the secure transmission of structured perception data from the traffic video private network to the traffic management private network. The dynamic holographic archive construction module connects to the traffic data interaction gateway, integrates multi-domain data, and achieves rapid and accurate association between people, vehicles and enterprises through three-dimensional association and hash matching algorithms. It designs a multi-factor dynamic weighting strategy, calculates the association weight according to the association level and business relevance, and then integrates the data value contribution to obtain the comprehensive weight. High-quality data is selected to directly update the archive, while low-quality data is stored in the verification queue. The multi-source risk assessment module matches model parameters according to real-time road network characteristics, integrates the outputs of multiple models such as time series prediction and spatial clustering, and generates early warning information containing risk level, cause and related files through a scenario priority weighting algorithm. It uses the handling feedback as a supervision signal to dynamically update feature weights and decision thresholds, and continuously improves the accuracy of assessment. The command and dispatch module presents the risk situation by overlaying early warning information, archive data and cross-network resource status. With the goal of optimal handling efficiency and minimal impact on the road network, it dynamically optimizes dispatch paths and resource allocation by combining real-time traffic flow and resource status, generates differentiated strategies, automatically triggers traffic video network monitoring and tracking, green wave timing of traffic signals, and synchronizes with traffic management network instructions to achieve seamless connection between analysis, dispatch and control.

2. The dynamic archiving and reliable updating system for heterogeneous traffic data according to claim 1, characterized in that, The specific methods for defining the traffic management private network and the traffic video private network are as follows: The traffic management private network refers to a dedicated network domain used by traffic management departments to carry core business systems and process structured business data. The traffic video private network refers to a dedicated network domain built independently of the traffic management private network, used to access front-end video image acquisition equipment, carry real-time video stream data, and perform video image analysis.

3. The dynamic archiving and reliable updating system for heterogeneous traffic data according to claim 1, characterized in that, The three-dimensional association in the dynamic holographic archive construction module is specifically as follows: Using license plate number, identity feature code, and enterprise feature code as association keys, a hash matching algorithm is used to quickly compare the license plate number captured by the traffic video network with the vehicle registration data of the traffic management network. If the match is successful, the identity feature code and enterprise feature code associated with the license plate number are extracted from the business database to form a temporary three-dimensional association relationship. If the match fails, the license plate number to be matched and its continuous spatiotemporal trajectory are stored in the waiting queue, and the similarity is calculated with the de-identified vehicle trajectory features in the historical archive, and the association relationship is dynamically updated.

4. The dynamic archiving and reliable updating system for heterogeneous traffic data according to claim 1, characterized in that, The multi-factor dynamic weighting strategy in the dynamic holographic archive construction module is specifically as follows: Set data weight evaluation factors, assign basic credibility weights to data from different sources, set data freshness weights that decay exponentially with the interval between the data generation time and the current time, calculate data integrity weights in reverse based on the proportion of missing fields, and assign data relevance weights based on the degree of association with the three-dimensional association key. The comprehensive weight of each data point is calculated using a linear weighted summation formula. Data with a comprehensive weight higher than a preset threshold are directly used for file updates, while data with a weight lower than the threshold are stored in a verification queue. The final decision on whether to include the data in the file is made after verification with subsequent related data.

5. The dynamic archiving and reliable updating system for heterogeneous traffic data according to claim 1, characterized in that, The model parameters are matched based on real-time road network characteristics. The specific method is as follows: The system pre-defines four core scenario model sets: key vehicle control, accident hazard warning, congestion prediction, and chain reaction of illegal events. Each model set contains multiple parameter templates adapted to different road network characteristics. It extracts road network feature parameters in real time, including area type, traffic flow level, time period attributes, weather conditions, and road construction status. The system uses a decision tree algorithm to classify real-time road network feature parameters and match them with the corresponding scenario model set. Based on the historical accuracy and real-time data freshness in that scenario, the system dynamically adjusts the model parameters through an adaptive weighted algorithm. The weights for historical accuracy and data freshness are reverse-corrected based on the assessment results in the most recent period, thus achieving dynamic optimization of the parameter templates.

6. The dynamic archiving and reliable updating system for heterogeneous traffic data according to claim 1, characterized in that, Early warning information is generated using a scene priority weighted algorithm. The specific method is as follows: A basic priority is assigned to each scenario model. The confidence level of each model's output is calculated based on the model's historical accuracy, input data completeness, and feature matching degree. The comprehensive priority of each model's output is calculated using a weighted formula of basic priority × confidence level. For conflicting judgment results, the result with the highest comprehensive priority is taken as the core warning conclusion. The core conclusion is combined with supplementary information from other model outputs to generate complete warning information.

7. The dynamic archiving and reliable updating system for heterogeneous traffic data according to claim 1, characterized in that, The specific method for dynamically optimizing scheduling paths and resource allocation is as follows: A scheduling decision model is constructed using reinforcement learning algorithms. Real-time traffic flow, police location and load, monitoring point coverage, and traffic signal control authority are used as the state space. A reward function is designed with response time, road network traffic efficiency loss, and police resource utilization rate as the core reward factors. The optimal scheduling path and resource combination scheme are explored through genetic algorithms to generate differentiated strategies.

8. The dynamic archiving and reliable updating system for heterogeneous traffic data according to claim 1, characterized in that, The method for automatically triggering traffic video network monitoring and tracking, and green wave timing of traffic signals is as follows: Based on the location and movement trajectory of the risk target in the early warning information, the Kalman filter algorithm is used to predict the target's route and the time of arrival at each intersection. Traffic video network monitoring points along the route are selected, and monitoring and tracking instructions are triggered in sequence according to the target's arrival time. At the same time, the monitoring images are transmitted back in real time. According to the predicted arrival time and real-time traffic flow, the start time and duration of the green light at each intersection are calculated using the green wave timing algorithm to generate a dynamic green wave scheme. The green wave scheme and the police dispatch route are issued synchronously and pushed to the traffic video network traffic signal controller through the trusted reverse security proxy channel of the traffic data interaction gateway. The timing parameters are dynamically fine-tuned according to the actual speed of the target to ensure smooth handling channels.

9. A dynamic archiving and reliable updating system for heterogeneous traffic data according to claim 8, characterized in that, Filter traffic video surveillance points along the travel route and trigger monitoring and tracking commands in sequence according to the target's arrival time. The specific method is as follows: Based on the predicted trajectory of the target and the spatial topology of the road network, monitoring points that cover the route and are unobstructed are selected. Priority is set according to the point resolution and monitoring angle. The points are sorted according to the estimated arrival time of the target. A time-series pre-triggering algorithm is adopted to preset the trigger threshold in advance, so as to achieve seamless tracking of points. If the target deviates from the trajectory, the points are re-selected in real time based on the latest position and the triggering order is dynamically adjusted.

Images