Methods and Systems for Massive Audit Data Storage and Processing Based on Private Cloud

By adopting a massive audit data storage and processing method based on private cloud, the problems of isolated and scattered data, difficulty in real-time collection and storage, and reliance on manual identification of toll evasion in the highway audit system have been solved. This method has achieved unified aggregation of cross-segment data, automated toll evasion identification, and data security, thereby improving the efficiency and security of the audit system.

CN123115252APending Publication Date: 2026-11-17GUIZHOU HONGXINDA HIGH SCI & TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611248598.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-18
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In the existing highway audit system, data is isolated and scattered, lacks a unified processing mechanism, and it is difficult to collect and store massive amounts of data in real time. The identification of toll evasion behavior relies on manual labor, which is not secure enough and makes it difficult to realize cross-segment trajectory reconstruction and automatic identification of various toll evasion behaviors.

Method used

This method for storing and processing massive audit data based on a private cloud collects real-time data from ETC gantries and toll stations, aggregates the data using Kafka message queues and TCP/JSON protocols, constructs a private cloud data platform with a primary and backup redundant structure, and combines road network models and various toll evasion behavior identification algorithms to achieve data standardization and trajectory reconstruction, generating a list of suspect vehicles and a chain of evidence.

Benefits of technology

It achieves unified aggregation and efficient storage of cross-road segment data, breaks through the limitations of a single data source, automatically identifies various toll evasion behaviors, improves audit efficiency and security, and ensures the continuity of data transmission and the confidentiality of sensitive information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115252A_ABST
    Figure CN123115252A_ABST
Patent Text Reader

Abstract

This invention relates to the field of audit data processing technology, particularly a method and system for storing and processing massive audit data based on a private cloud. The advantages of this invention are: it constructs an end-to-end data storage and processing architecture from ETC gantries and toll station collection equipment to the provincial data aggregation center, and then via fiber optic dedicated lines and front-end firewalls to the private cloud data platform; it utilizes Kafka message queues to distribute data to different topics according to data type and combines TCP communication protocol and JSON message protocol for real-time push; and it employs a primary / backup link redundancy structure to ensure data transmission continuity. This effectively solves the problems of isolated and scattered data in existing technologies, the inability to aggregate data across road segments, and the easy backlog and loss of massive high-concurrency data during collection and transmission. It enables traffic data from different road segments and collection points to be uniformly aggregated to the private cloud data platform, providing a complete and reliable data foundation for subsequent full-network vehicle trajectory reconstruction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of audit data processing technology, and in particular to a method and system for storing and processing massive audit data based on a private cloud. Background Technology

[0002] With the complete elimination of provincial toll stations on national highways in 2020, my country's highways achieved true nationwide networked toll collection. Following the removal of provincial toll stations, a large number of ETC gantry systems were installed, leading to an exponential increase in highway traffic data. It is estimated that the national ETC gantry system generates hundreds of millions of transaction records daily. At the same time, problems such as difficulties in ETC vehicle supervision, toll collection verification, and auditing of toll evasion have become increasingly prominent. Traditional auditing methods are ill-suited to the new toll collection environment. With the widespread application of ETC technology, toll transaction data has experienced explosive growth and diverse types, making manual review not only inefficient but also prone to oversights. The wider scope of highway toll evasion, the massive volume of audit data, and the complexity of data analysis models have become increasingly prominent technical challenges that data auditing management must address.

[0003] Currently, most aspects of highway operation and management, including networked toll collection management, toll audit management, and toll monitoring management, have been digitized. However, each system exists as an information silo, often unable to be interconnected. In particular, toll data is disconnected from monitoring audio and video information, making it difficult to effectively narrow down the scope of audit information from massive amounts of raw toll data. Data from different systems, such as video surveillance, ETC gantries, and weather sensors, uses different formats and transmission protocols, hindering effective data sharing and collaborative analysis. Traffic data and images from ETC gantries are scattered across the gantry servers in the respective toll station's computer room. Operating units on different road sections can only query data within their own jurisdiction, making it difficult to fully reconstruct vehicle trajectories across road sections and provinces. Furthermore, a large amount of information relies on manual identification, making it difficult to identify key information and reconstruct the chain of evidence when faced with massive amounts of data and various new types of toll evasion.

[0004] In terms of data storage and processing, the massive, high-concurrency, and time-sensitive "data deluge" is impacting existing storage architectures. Existing systems lack unified standards and efficient processing mechanisms in data collection, transmission, and storage, failing to meet the timeliness requirements of real-time auditing. Traffic data and images from ETC gantries in various provinces are stored in a scattered manner. While highway operators can connect to corresponding servers at their respective road section centers for querying and image retrieval, frequent and proactive connections to server databases disrupt the operation of existing ETC gantry tolling programs and data uploads. Vehicle travel trajectories are not necessarily fixed within a specific road section or area; when trajectories exceed jurisdictional boundaries, they become beginningless or endingless, making cross-provincial backend path reconstruction difficult.

[0005] Regarding the identification of toll evasion, the methods used are constantly evolving. Illegal activities have progressed from simple methods like card swapping and vehicle reversal to using high-tech equipment to interfere with ETC signals. Traditional auditing methods mainly rely on manual comparison of each toll transaction record, which is difficult to screen and lacks supporting images and traces, resulting in incomplete screening of abnormal data. Current technologies lack a unified identification algorithm model system for various types of toll evasion, making it impossible to systematically and automatically identify and classify different types of toll evasion behaviors such as inconsistent vehicle identification tags, masked media, large vehicles with small tags, media swapping, and underpayment.

[0006] Furthermore, existing auditing systems have shortcomings in data security. Auditing data involves sensitive information such as vehicle access IDs, license plate numbers, toll fees, and rates, and current technology lacks effective encryption protection mechanisms for this sensitive data. A unified encrypted storage scheme for sensitive data has not yet been established across the entire data storage and processing chain, posing a security risk of data leakage.

[0007] In summary, existing highway audit data processing technologies have at least the following shortcomings: 1. Data from various systems is isolated and scattered, lacking a unified data aggregation and standardized processing mechanism; 2. The real-time collection, transmission, and storage of massive amounts of traffic data lack efficient and reliable architectural support; 3. Vehicle trajectory reconstruction relies on a single data source, making it difficult to overcome the limitations of license plate recognition in cases of license plate forgery or obstruction; 4. There is a lack of systematic identification algorithm models for various types of toll evasion behaviors. Toll evasion identification relies on manual screening, which is inefficient and has limited coverage. 5. Audit data lacks effective security encryption protection during storage and processing. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for storing and processing massive audit data based on private cloud, which effectively solves the deficiencies of the prior art.

[0009] The objective of this invention is achieved through the following technical solution: a method for storing and processing massive audit data based on a private cloud, comprising the following steps: S1: Real-time collection of basic vehicle traffic data is achieved through data collection equipment deployed at ETC gantries and toll stations on highways. The basic traffic data includes ETC gantry identification data, ETC gantry transaction data, cash transaction data at toll station entrances, non-cash transaction data at toll station entrances, cash transaction data at toll station exits, and non-cash transaction data at toll station exits. S2: The basic traffic data collected in step S1 is aggregated to the provincial data aggregation center through the toll collection network. The provincial data aggregation center then distributes the data to different topics according to data type through the Kafka message queue, and pushes it in real time using TCP as the communication protocol and JSON as the message protocol. S3: The operation and maintenance service center connects to the provincial data aggregation center's toll network through a leased fiber optic line, receives data pushed by the Kafka message queue through a front-end firewall and a front-end forwarding server, and stores the data in a private cloud data platform. The private cloud data platform adopts a primary and backup link redundancy structure. S4: Standardize the data stored in step S3, unify the data format and encoding, and generate standardized common data; S5: Construct a weighted directed graph-based road network model based on highway network parameters. The road network parameters include toll road information, toll section information, toll unit information, toll station information, toll gantry information, road network node coding information, and adjacent node relationship information. Highway toll stations and ETC gantries are parsed as nodes of the graph, and reachable paths between any two ordered nodes are parsed as edges of the graph. S6: Input the standardized traffic data from step S4 into the road network model from step S5, and combine it with vehicle traffic medium information and license plate recognition information to reconstruct the vehicle's driving trajectory and generate the complete vehicle driving trajectory. S7: Based on the vehicle driving trajectory restored in step S6, call the pre-configured multiple toll evasion behavior recognition algorithm models to analyze the trajectory, identify abnormal behaviors and mark abnormal types, and generate a list of suspect vehicles and corresponding evidence chain data. S8: Push the list of suspected vehicles and evidence chain data generated in step S7 to the audit business front desk for auditors to manually review and confirm, and update the vehicle status based on the review results.

[0010] Preferably, in step S2, the data is distributed to different topics according to the data type as follows: ETC gantry identification data is distributed to the first topic, ETC gantry transaction data is distributed to the second topic, cash transaction data at the toll station exit is distributed to the third topic, non-cash transaction data at the toll station exit is distributed to the fourth topic, cash transaction data at the toll station entrance is distributed to the fifth topic, and non-cash transaction data at the toll station entrance is distributed to the sixth topic.

[0011] Preferably, in step S7, the multiple toll evasion behavior identification algorithm models include at least the following: vehicle tag mismatch and OBU misuse identification model, U-turn identification model, abnormal route identification model, shielded medium identification model, one vehicle with multiple media identification model, large vehicle with small tag identification model, medium swapping identification model, and underpayment identification model.

[0012] Preferably, the discrepancy in vehicle identification tags and the application of the OBU identification model are identified through the following steps: 1) Compare the license plates collected from the toll station entry and exit data with the license plates in the toll medium, and cross-compare the license plates identified in the ETC gantry license plate recognition data with the license plates of the toll vehicles in the ETC gantry transaction data. Filter out vehicle information with license plates but no payment and vehicles with payment but no license plates and add it to the pending list. 2) Aggregate the complete passage trajectory of the vehicle by the passage identification ID. When the license plate data and transaction data do not match on 5 or more consecutive ETC gantries, and the similarity between the license plate trajectory and the transaction trajectory is higher than 90%, mark the vehicle trajectory as vehicle license plate mismatch / using OBU and output suspected real license plate.

[0013] Preferably, in any of the above schemes, the shielding medium identification model identifies anomalies through the following steps: 1) Based on the passage identifier ID in the ETC gantry transaction data, associate all ETC gantries for the vehicle's passage to form a list of trajectory points, and filter out the trajectory nodes in the ETC gantry transaction data where the license plate of the tolling vehicle is missing and add them to the list to be processed; 2) Based on the start and end times of the billing transaction in the ETC gantry transaction data of this passage, the node range is narrowed down, and the license plate recognition data of the vehicle in the ETC gantry within this range is automatically compared to form a complete path. When the passage meets the condition that there is license plate recognition data but no transaction data in 5 consecutive ETC gantries, the vehicle trajectory is marked as a shielding medium.

[0014] Preferably, the large vehicle small label recognition model identifies anomalies through the following steps: 1) Vehicle information is synchronized from the provincial vehicle information database to this system, or the highway owner enters or modifies vehicle information in this system; 2) Match the license plate number and color based on the license plate in the toll station entry and exit data and the license plate of the tolling vehicle in the ETC gantry transaction data. When the vehicle type and number of axles in the toll station entry and exit data, the number of axles and the entry vehicle type in the ETC gantry transaction data are inconsistent with the number of axles and the vehicle type in the vehicle information database of this system, mark the vehicle trajectory as a large vehicle with a small label.

[0015] Preferably, in any of the above schemes, the underpayment identification model calculates the amount of evaded payment through the following steps: 1) Using the abnormal driving trajectory reconstructed by the various toll evasion behavior identification algorithm models, combined with the ETC gantry toll rate table information, the theoretical toll is calculated segment by segment according to the gantry sequence through which the trajectory passes; 2) Compare the theoretical toll with the actual amount collected in the toll station exit data. If the actual amount collected is less than the theoretical toll, it is marked as underpayment. The difference is calculated as suspected loss toll and an audit evidence chain including trajectory path, toll calculation details and exit transaction records is generated.

[0016] Preferably, in step S3, the private cloud data platform performs SM2 encryption on the vehicle's access ID, license plate number, toll fee, and rate data.

[0017] A massive audit data storage and processing system based on a private cloud includes: The data acquisition module is deployed at highway ETC gantries and toll stations to collect basic vehicle traffic data in real time. The basic traffic data includes ETC gantry identification data, ETC gantry transaction data, toll station entrance cash transaction data, toll station entrance non-cash transaction data, toll station exit cash transaction data, and toll station exit non-cash transaction data. The data aggregation and distribution module is deployed in the provincial data aggregation center. It is used to receive the basic traffic data collected by the data collection module through the toll collection network, and distribute it to different topics according to data type through the Kafka message queue. It uses TCP as the communication protocol and JSON as the message protocol for real-time push. The data receiving and storage module, deployed in the operation and maintenance service center, includes a fiber optic leased line access unit, a front-end firewall unit, a front-end forwarding server unit, and a private cloud storage unit. It is used to access the toll network of the provincial data aggregation center through a leased fiber optic leased line, receive data pushed by the Kafka message queue through the front-end firewall and the front-end forwarding server, and store the data in the private cloud data platform. The private cloud data platform adopts a primary and backup link redundancy structure. The data standardization module is used to standardize the data stored in the data receiving and storage module, unify the data format and encoding, and generate standardized common data. The road network model construction module is used to construct a weighted directed graph form of road network model based on highway road network parameters. The road network parameters include toll road information, toll section information, toll unit information, toll station information, toll gantry information, road network node coding information, and adjacent node relationship information. The module resolves highway toll stations and ETC gantries as nodes of the graph and resolves reachable paths between pairs of ordered nodes as edges of the graph. The trajectory restoration module is used to input the standardized traffic data generated by the data standardization module into the road network model constructed by the road network model construction module, and combine the vehicle traffic medium information and license plate recognition information to restore the vehicle's driving trajectory and generate the complete vehicle driving trajectory. The abnormal behavior identification module is used to analyze the trajectory based on the vehicle driving trajectory restored by the trajectory restoration module, call a variety of pre-configured toll evasion behavior identification algorithm models, identify abnormal behaviors and mark abnormal types, and generate a list of suspect vehicles and corresponding evidence chain data. The audit business processing module is used to push the list of suspected vehicles and evidence chain data generated by the abnormal behavior identification module to the audit business front end for auditors to manually review and confirm, and update the vehicle status based on the review results.

[0018] Preferably, any of the above solutions also includes: The vehicle control module is used to monitor vehicles on the control list in real time. When a vehicle on the control list passes through the ETC gantry, it pushes the latest location information of the vehicle to the auditors. The toll collection work order management module is used to generate toll collection work orders based on manually verified vehicles that have evaded tolls, push the toll collection work orders to the toll collection work order system, receive the review results returned by the toll collection work order system, and update the vehicle blacklist status synchronously. The private cloud storage unit is also used to store vehicle access identification ID, license plate number, toll fee and rate data using the national cryptographic standard SM2.

[0019] The present invention has the following advantages: 1. This method and system for storing and processing massive audit data based on a private cloud constructs an end-to-end data storage and processing architecture, from ETC gantries and toll station collection equipment to the provincial data aggregation center, and then through fiber optic dedicated lines and front-end firewalls to the private cloud data platform. It utilizes Kafka message queues to distribute data to different topics according to data type and combines TCP communication protocol and JSON message protocol for real-time push. At the same time, it adopts a primary and backup link redundancy structure to ensure the continuity of data transmission. It effectively solves the problems of isolated and scattered data in various systems, inability to aggregate cross-road segment data, and easy backlog and loss of massive high-concurrency data in the collection and transmission process in existing technologies. It enables traffic data from different road segments and different collection points to be uniformly aggregated to the private cloud data platform, providing a complete and reliable data foundation for subsequent full-network vehicle trajectory reconstruction.

[0020] 2. This method and system for storing and processing massive audit data based on a private cloud resolves toll stations and ETC gantries as nodes in a graph and resolves reachable paths between any two ordered nodes as edges in the graph. Then, in step S6, standardized traffic data is input into the road network model and combined with traffic medium information and license plate recognition information to reconstruct the trajectory. This makes the reconstruction of vehicle driving trajectory no longer dependent on a single data source. Even in the case of license plate forgery, obstruction, or recognition failure, the vehicle position can still be inferred and completed through traffic medium information and road network topology constraints. This breaks through the limitation of traditional reliance on license plate recognition in the case of abnormal situations, and realizes the accurate reconstruction of the vehicle's real driving trajectory.

[0021] 3. This method and system for storing and processing massive audit data based on a private cloud calls pre-configured multiple toll evasion behavior identification algorithm models to analyze the restored vehicle driving trajectory. It integrates the identification logic of various toll evasion behaviors such as vehicle tag discrepancies, shielded media, large vehicle with small tag, media swapping, and underpayment into a unified algorithm model system. This enables the system to automatically match the corresponding detection rules and mark the abnormality type for different types of abnormal driving patterns. It overcomes the shortcomings of existing technologies that lack a systematic toll evasion identification model system and rely on manual comparison of transaction data, resulting in low screening efficiency and limited coverage. It realizes automated and systematic identification of diverse toll evasion behaviors and simultaneously generates a complete list of suspected vehicles and evidence chain data.

[0022] 4. This method and system for storing and processing massive audit data based on a private cloud uses a private cloud data platform to encrypt and store toll identification IDs, license plate numbers, toll fees, and rates using the national cryptographic standard SM2. Combined with access control policies of a front-end firewall, it securely isolates data transmission between the toll collection network and the operation and maintenance service center, ensuring that sensitive audit data remains encrypted throughout the entire storage chain. Even if the storage medium is illegally accessed, plaintext information cannot be directly obtained, effectively preventing the risk of audit data leakage during storage and processing, and ensuring the security and compliance of key data in toll collection audit business.

[0023] 5. This method and system for storing and processing massive amounts of audit data based on a private cloud pushes the list of suspected vehicles and evidence chain data to the audit business front end for manual review. Combined with the vehicle control module, it monitors the vehicles on the control list in real time and pushes the latest location information to the auditors when the vehicles pass through the ETC gantry. At the same time, the toll payment work order management module generates toll payment work orders for confirmed toll evasion vehicles and pushes them to the toll payment work order system, and receives the review results to update the blacklist status synchronously. This forms a business closed loop from anomaly identification, manual review, real-time control to toll payment collection and blacklist synchronization, which transforms the audit work from passive retrospective to active intervention in the process, improving the timeliness and success rate of toll evasion collection.

[0024] 6. This method and system for storing and processing massive audit data based on a private cloud adopts a fiber optic leased line access scheme with a primary and backup redundant link structure, combined with a two-level security isolation mechanism of a front-end firewall and a front-end forwarding server. This allows data transmission between the toll collection network and the operation and maintenance service center to seamlessly switch to the backup link when the primary link fails. At the same time, the firewall's access control policies impose fine-grained restrictions on source addresses, destination addresses, and business ports, ensuring both the accessibility of the audit system for vehicle image analysis and evidence collection in the toll collection network nodes and achieving secure isolation of the toll collection network. This solves the problem in existing technologies where frequent connections to the gantry server to query data affect the operation of the billing program and restrict cross-road segment trajectory queries, ensuring the independent and stable operation of the toll collection business system and the audit system. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the system framework of the present invention. Detailed Implementation

[0026] The present invention will be further described below with reference to the accompanying drawings, but the scope of protection of the present invention is not limited to the following description.

[0027] Example 1: As Figure 1 As shown, this embodiment provides a method for storing and processing massive audit data based on a private cloud. This method acquires vehicle traffic data in real time through data collection devices deployed at highway ETC gantries and toll stations. Through steps such as data aggregation, distribution, storage, standardization, road network modeling, trajectory reconstruction, anomaly identification, and audit processing, it achieves intelligent identification of highway toll evasion. It includes the following steps: Step S1: Data Acquisition.

[0028] Data collection equipment is deployed at ETC gantries and toll stations on highways to collect basic vehicle traffic data in real time. Specifically, at ETC gantries, gantry identification data and gantry transaction data are collected. Gantry identification data includes gantry number, identified license plate, vehicle type, passage time, lane number, image storage address, and recognition confidence level. Gantry transaction data includes gantry number, OBU device ID, ETC card number, toll vehicle license plate, toll transaction time, transaction amount, and mileage. Cash transaction data and non-cash transaction data are collected at toll station entrances and exits, respectively. Entrance data includes toll station number, lane number, license plate number, vehicle type, cash amount or electronic payment transaction serial number, passage time, and toll station employee number. Exit data similarly includes toll station number, lane number, license plate number, vehicle type, payment method, transaction serial number, transaction amount, and passage time. All data collection equipment maintains a real-time communication connection with the provincial data aggregation center via a dedicated toll network.

[0029] Step S2: Data aggregation and distribution.

[0030] The collected basic traffic data is aggregated to the provincial data aggregation center via a dedicated toll collection network. The provincial data aggregation center deploys a Kafka cluster, distributing data to different topics according to data type. Specifically, ETC gantry identification data is distributed to the first topic (gantry identification data), ETC gantry transaction data to the second topic (gantry transaction data), toll station exit cash transaction data to the third topic (exit cash transaction data), toll station exit non-cash transaction data to the fourth topic (exit non-cash transaction data), toll station entrance cash transaction data to the fifth topic (entrance cash transaction data), and toll station entrance non-cash transaction data to the sixth topic (entrance non-cash transaction data). Data is transmitted via a persistent TCP connection and encapsulated in JSON. Each message contains a header and a body. The header includes a msgId for deduplication verification, a msgType to identify the data type, and a timestamp to record the generation time. The body contains specific business fields. The Kafka cluster employs a partitioning strategy, using msgType hashes for partitioning. Delivery semantics are configured with AtLeastOnce to ensure no data loss, and GZIP compression is enabled to improve transmission efficiency. End-to-end transmission latency is controlled within 500ms, and system throughput is no less than 5000 messages per second.

[0031] Step S3: Data reception and storage.

[0032] The operations and maintenance service center is located in a different physical location from the provincial data aggregation center, separated by a municipal road and lacking a dedicated high-speed communication system. Therefore, the operation and maintenance service center leases two dedicated fiber optic lines from a carrier to access the provincial data aggregation center's toll collection network. These two lines employ a primary / backup redundant structure, each with a bandwidth of 100Mbps. When the primary link is functioning normally, the backup link is in hot standby mode, allowing seamless switching to the backup link in case of a primary link failure. A front-end firewall is installed between the toll collection network and the operation and maintenance service center, strictly enforcing ACL access control policies, specifying source and destination addresses and service ports, and implementing least privilege access according to the policy optimization principle. Data received via the firewall and front-end forwarding server is written to the distributed storage system within the private cloud data platform. The private cloud data platform is deployed in the operation and maintenance service center's server room, built using a server cluster, providing highly available and scalable storage services. Meanwhile, the private cloud data platform uses the national cryptographic SM2 algorithm to encrypt and store sensitive information such as vehicle access ID, license plate number, toll fee and rate data. The encryption key is kept by an independent key management system to ensure that the data remains encrypted throughout the entire storage chain and prevent data leakage.

[0033] Step S4: Data standardization.

[0034] The data stored in step S3 is standardized. Since the traffic data from different sources differ in field naming, data format, and encoding methods—for example, the vehicle passage time field in gantry license plate data is named `passTime`, and the time field in toll station transaction data is named `transTime`—standardization unifies the data from different sources to the same field naming convention. A unified time format uses UTC + 8 millisecond timestamps. Unified license plate color encoding includes nine values: blue, yellow, black, white, gradient green, yellow-green, blue-white gradient, and undetermined. Unified vehicle type classification encoding includes eight categories, from Class I passenger vehicles to Class IV freight vehicles. Standardized traffic data is generated for upper-layer applications to subscribe to and use.

[0035] Step S5: Road network model construction.

[0036] A weighted directed graph model of the highway network was constructed based on highway network parameters. These parameters were obtained from the Guizhou Provincial Transportation Comprehensive Administrative Law Enforcement Technology and Information Support Center, including toll road information (owner number and owner unit name), toll segment information (segment number, segment name, and owner), toll unit information, toll station information (toll station number, toll station name, latitude and longitude, and the assigned road segment), toll gantry information (gantry number, gantry name, latitude and longitude, and the assigned road segment), network node coding information, and adjacent node relationship information (adjacent node pairs, directions, and distances). Based on these parameters, highway toll stations and ETC gantries were resolved as nodes in a weighted directed graph, and reachable paths between any two ordered nodes were resolved as edges in the graph, with the edge weight representing the actual network distance between nodes. This weighted directed graph model accurately reflects the topological structure and physical connections of the highway network, providing spatial constraints and a basis for subsequent trajectory reconstruction and path estimation.

[0037] Step S6: Trajectory reconstruction.

[0038] The standardized traffic data from step S4 is input into the road network model in step S5. Combined with vehicle traffic medium information and license plate recognition information, the vehicle's trajectory is reconstructed. Specifically, all trajectory points for a single vehicle passage are aggregated by traffic identifier ID, including toll transaction location points in ETC gantry transaction data, license plate recognition location points in ETC gantry license plate recognition data, and the locations of toll station entrances and exits. When trajectory points are missing, such as when some gantries only have license plate recognition without transactions or only transactions without license plate recognition, the adjacent relationships and distance information between nodes in the road network model, combined with the passage time sequence, are used to spatially fill in the missing nodes. The license plate recognition trajectory is then matched with the transaction trajectory to generate a continuous and complete vehicle trajectory. In this way, even if the license plate is forged, obscured, or recognition fails, the vehicle's true location can still be inferred through traffic medium information and road network topology constraints, overcoming the limitations of traditional reliance solely on license plate recognition.

[0039] Step S7: Abnormal behavior identification.

[0040] Based on the vehicle trajectory reconstructed in step S6, various pre-configured toll evasion behavior identification algorithm models are invoked to analyze the trajectory. In this embodiment, the pre-configured toll evasion behavior identification algorithm models include a vehicle tag mismatch and OBU misuse identification model, a U-turn identification model, a path anomaly identification model, a shielded medium identification model, a vehicle with multiple media identification model, a large vehicle with a small tag identification model, a medium swapping identification model, and a low-payment identification model. Each model runs in parallel, performing multi-dimensional anomaly detection on each reconstructed trajectory, identifying abnormal behaviors and marking anomaly types, and generating a list of suspect vehicles and corresponding evidence chain data.

[0041] The identification process for vehicle license plate mismatch and OBU misuse is as follows: First, the license plate data collected from the toll station entry and exit data is compared with the license plates in the toll medium. The license plates identified in the ETC gantry license plate data are cross-compared with the license plates of the toll vehicles in the ETC gantry transaction data. Vehicle information with license plates but no transactions (i.e., license plate data exists but there is no corresponding transaction data) and vehicles with transactions but no license plates (i.e., transaction data exists but there is no corresponding license plate data) are added to the pending list. Second, the complete passage trajectory of the vehicle is aggregated by the toll identification ID. When the license plate data and transaction data do not match on 5 or more consecutive ETC gantries, and the similarity between the license plate trajectory and the transaction trajectory calculated based on the longest common subsequence algorithm is higher than 90%, it is determined that there is suspicion of OBU misuse. The vehicle trajectory is marked as vehicle license plate mismatch or OBU misuse and a suspected real license plate is output.

[0042] The identification process of the shielded medium identification model is as follows: First, based on the passage identifier ID in the ETC gantry transaction data, all ETC gantries for the vehicle's passage are associated to form a list of trajectory points. Trajectory nodes with missing license plates in the ETC gantry transaction data are filtered out and added to the pending list. Second, the node range is narrowed down based on the start and end times of the toll transaction in the ETC gantry transaction data for the passage. The license plates in the ETC gantry license plate recognition data of the vehicle within this range are automatically compared to form a complete path. When the passage meets the condition that there is license plate recognition data but no transaction data in 5 consecutive ETC gantries, it is determined that the vehicle's passage medium is shielded, and the vehicle trajectory is marked as shielded medium.

[0043] The identification process of the large vehicle with small license plate identification model is as follows: First, vehicle information is synchronized from the provincial vehicle information database to this system, or the highway owner enters or modifies vehicle information in this system; Second, the license plate number and color are matched based on the license plate in the toll station entry and exit data and the license plate of the tolling vehicle in the ETC gantry transaction data. When the vehicle type and axle count in the toll station entry and exit data, the axle count and entry vehicle type in the ETC gantry transaction data are inconsistent with the axle count and vehicle type in this system's vehicle information database, it is determined that there is a suspicion of vehicle type mismatch, and the vehicle trajectory is marked as large vehicle with small license plate.

[0044] The identification process of the underpayment identification model is as follows: First, the abnormal driving trajectory is reconstructed using the various toll evasion behavior identification algorithm models. Combined with the ETC gantry rate table information, the theoretical toll is calculated segment by segment according to the gantry sequence traversed by the trajectory. The theoretical toll is equal to the sum of the product of the rate corresponding to each gantry segment and the mileage of that segment. The mileage is determined based on the actual road network distance pre-stored between nodes in the road network model. Second, the theoretical toll is compared with the actual amount collected in the toll station exit data. When the actual amount collected is less than the theoretical toll, it is marked as underpayment, and the difference is calculated as the suspected lost toll. An audit evidence chain containing the trajectory path, toll calculation details, and exit transaction records is generated.

[0045] Step S8: Audit business processing.

[0046] The suspected vehicle list and evidence chain data generated in step S7 are pushed to the audit front-end. Auditors view the detailed trajectory information, anomaly type markers, and related evidence images of the suspected vehicles on the front-end for manual review and confirmation. Vehicles confirmed to have evaded tolls are updated to a "recovery" status; vehicles confirmed to have no false alarms have their anomaly markers cleared. Simultaneously, a toll collection work order is generated for confirmed toll-evading vehicles. The work order includes information such as vehicle license plate, actual vehicle model, permitted vehicle type, license plate color, number of axles, submitting unit number, and a list of abnormal vouchers, and is pushed to the toll collection work order system via HTTP interface in zip compressed format. After reviewing and approving the work order, the toll collection work order system returns the evaluation result to this system, which updates the vehicle's recovery status and blacklist status accordingly. For vehicles included in the blacklist, the system monitors them in real time through the vehicle deployment module. When a vehicle on the deployment list passes any ETC gantry, the system pushes the latest vehicle location information to the auditors, enabling online real-time tracking.

[0047] In this embodiment, in step S3, the private cloud data platform encrypts and stores the vehicle's access ID, license plate number, toll fee, and rate data using the national cryptographic algorithm SM2. Specifically, the encryption method is as follows: the asymmetric encryption algorithm SM2 is used to encrypt the sensitive field values ​​with a system-level public key before storage. Decryption is performed using the corresponding private key. The encryption and decryption operations are uniformly scheduled by an independent key management service, and the keys are rotated periodically to ensure the confidentiality and integrity of sensitive data throughout the entire storage chain.

[0048] Example 2: Figure 2 As shown, this embodiment provides a massive audit data storage and processing system based on a private cloud, used to implement the method described in Embodiment 1. The system adopts a modular architecture, and completes the entire business process from data collection to audit processing through the collaborative work of various functional modules.

[0049] It includes a data acquisition module, a data aggregation and distribution module, a data receiving and storage module, a data standardization module, a road network model construction module, a trajectory restoration module, an abnormal behavior identification module, an audit business processing module, a vehicle control module, and a toll collection work order management module.

[0050] The data acquisition module is deployed at highway ETC gantries and toll stations. At the ETC gantries, the module includes a license plate recognition subunit and a transaction acquisition subunit, which collect gantry license plate recognition data and gantry transaction data, respectively. Gantry license plate recognition data includes gantry number, recognized license plate, vehicle type, passage time, lane number, image storage address, and recognition confidence level. Gantry transaction data includes gantry number, OBU device ID, ETC card number, toll vehicle license plate, toll transaction time, transaction amount, and mileage. At the toll stations, the module includes entrance cash transaction acquisition subunit, entrance non-cash transaction acquisition subunit, exit cash transaction acquisition subunit, and exit non-cash transaction acquisition subunit, which collect transaction data for each type of toll station. Entrance cash transaction data includes toll station number, lane number, license plate number, vehicle type, cash amount, passage time, and toll station employee number. Entrance non-cash transaction data includes toll station number, lane number, license plate number, vehicle type, payment method, transaction serial number, transaction amount, and passage time. Exit cash transaction data and exit non-cash transaction data similarly contain their respective fields. Each data collection subunit communicates with the provincial data aggregation center through a dedicated toll network and uploads data in real time with an upload interval of no more than 5 seconds.

[0051] The data aggregation and distribution module is deployed at the provincial data aggregation center. This module includes a Kafka cluster management unit, which receives basic toll data uploaded by the data collection module via the toll collection network and distributes the data to different Kafka Topics according to data type. Specifically, six topics are created in the Kafka cluster: the first topic stores ETC gantry identification data, the second topic stores ETC gantry transaction data, the third topic stores toll station exit cash transaction data, the fourth topic stores toll station exit non-cash transaction data, the fifth topic stores toll station entrance cash transaction data, and the sixth topic stores toll station entrance non-cash transaction data. Data is transmitted via TCP connections and encapsulated using JSON as the message protocol. The Kafka cluster is deployed with three nodes, twelve partitions, a replication factor of two, and a message retention time of 72 hours to ensure that consumption can resume from the point of interruption after a brief failure of the audit system.

[0052] The data receiving and storage module is deployed in the operations and maintenance service center. This module includes a fiber optic leased line access unit, a front-end firewall unit, a front-end forwarding server unit, and a private cloud storage unit. The fiber optic leased line access unit leases two 100Mbps leased lines from a carrier, using a primary / backup redundant structure to access the provincial data aggregation center's toll network. When the primary link fails, it automatically switches to the backup link, with a switching time of no more than 3 seconds. The front-end firewall unit is located between the toll network and the operations and maintenance service center, enforcing ACL access control policies, specifying source addresses, destination addresses, and service ports, allowing necessary traffic according to the principle of least privilege, and blocking unauthorized access. The front-end forwarding server unit receives Kafka data streams filtered by the firewall and performs protocol conversion and format verification. The private cloud storage unit consists of multiple servers forming a distributed storage cluster, employing a master-slave replication architecture. Data is synchronously replicated to at least two replica nodes during writing to ensure high data availability. The private cloud storage unit is also equipped with the national cryptographic SM2 encryption engine, which encrypts and stores vehicle access ID, license plate number, toll and rate data. Sensitive data is persisted in ciphertext form to prevent plaintext information from being leaked when the storage medium is illegally accessed.

[0053] The data standardization module connects to the data receiving and storage module, reading the stored raw data and performing standardization processing. This module includes a format conversion unit, a field mapping unit, and an encoding unification unit. The format conversion unit converts time fields from different data sources to a unified UTC+8 milliseconds timestamp format; the field mapping unit maps fields with the same business meaning from different data sources to unified field names, for example, mapping the passage time field in license plate recognition data and the transaction time field in transaction data to a unified standard time field; the encoding unification unit unifies license plate color codes, vehicle type classification codes, payment method codes, etc., into a standard internal encoding table. The standardized data is then written back to the private cloud storage unit for use by upper-layer modules.

[0054] The road network model construction module obtains highway network parameters from the provincial technical information center, including toll road information, toll section information, toll unit information, toll station information, toll gantry information, road network node coding information, and adjacent node relationship information. This module resolves toll stations and ETC gantries as nodes in a weighted directed graph, and resolves reachable paths between any two ordered nodes as edges in the graph, with the edge weight being the actual road network distance between the nodes. A complete road network topology graph is constructed and stored in a private cloud storage unit. The road network model supports periodic incremental updates; when new gantries or toll stations are added, the model can be updated incrementally without requiring a full reconstruction.

[0055] The trajectory reconstruction module is connected to both the data standardization module and the road network model construction module. This module includes a trajectory point association unit, a spatial point completion unit, and a path fitting unit. The trajectory point association unit aggregates all trajectory points for a single vehicle passage based on the passage identifier ID, including transaction location points and license plate location points. When a missing trajectory point is detected, the spatial point completion unit uses the node adjacency relationships in the road network model, combined with the passage time sequence, to infer and complete the missing location. The path fitting unit matches the completed trajectory point sequence with the paths in the road network model to generate a continuous and complete vehicle travel trajectory. The trajectory reconstruction results are stored in a private cloud storage unit indexed by the trajectory ID, supporting retrieval based on multiple conditions such as license plate, time, and location.

[0056] The abnormal behavior identification module is connected to the trajectory reconstruction module. This module is pre-configured with eight toll evasion behavior identification algorithm models, including models for vehicle license plate mismatch and OBU misuse, U-turn identification, abnormal route identification, shielded medium identification, one vehicle with multiple media identification, large vehicle with small license plate identification, medium swapping identification, and underpayment identification. Each algorithm model executes in parallel, and each reconstructed trajectory is sequentially detected by each model. The detection logic of each model is consistent with the description in step S7 of Example 1. The abnormal behavior identification module outputs a list of suspect vehicles. Each suspect record includes the vehicle license plate, abnormality type, time interval of the abnormality, list of involved ETC gantries, confidence score, and evidence chain data index. The evidence chain data includes vehicle license plate information, driving route data, and ETC gantry captured image data, which are automatically collected and stored by each algorithm model during execution.

[0057] The audit business processing module is connected to the abnormal behavior identification module. This module includes a work order push unit, a review management unit, and a status synchronization unit. The work order push unit assembles the information of confirmed toll evaders into a toll payment work order. The work order includes information such as vehicle license plate, actual vehicle type, permitted vehicle type, license plate color, number of axles, submitting unit number, and a list of abnormal vouchers. After being encapsulated in JSON format, it is pushed to the toll payment work order system via an HTTP POST interface. The review management unit receives the manual review results from auditors at the business front end and supports operations such as approving, rejecting, and marking suspected records for investigation. The status synchronization unit receives the review results returned by the toll payment work order system and updates the corresponding vehicle status field in its system accordingly.

[0058] The vehicle control module provides list management and real-time tracking functions. The list management subunit supports adding, deleting, modifying, and querying vehicles on the control list, which includes blacklisted vehicles and vehicles under special monitoring. The real-time tracking subunit is connected in real-time to the gantry transaction data topic in the Kafka cluster. When a vehicle on the control list passes through any ETC gantry, this subunit matches the vehicle to the list the moment it receives the gantry transaction data and immediately pushes the latest vehicle location information to auditors via in-system message alerts and SMS notifications. The supplementary payment work order management module is connected to the audit business processing module. This module includes a work order generation unit, a work order push unit, and a result synchronization unit. The work order generation unit generates a work order data package that meets the format requirements according to the interface specifications of the supplementary payment work order system, based on the suspected records output by the abnormal behavior identification module and the review results of the auditors. The work order data package is a zip compressed file containing a workOrder.json file with a fixed name and multiple folders named after credential IDs. Each folder stores the corresponding evidence images. The work order push unit uploads the work order data package to the supplementary payment work order system via an HTTP interface. The result synchronization unit receives the review results from the supplementary payment work order system through a callback interface, parses the review status and review comments, synchronizes the blacklist review status to the system database, and updates the recovery status field of the corresponding vehicle.

[0059] Preferably, in this embodiment, the modules communicate via RESTful APIs. The backend is developed using Java and Spring Boot 3.x frameworks, the frontend uses Vue 3.2 framework, the database uses MySQL 8 to store business data and configuration information, ClickHouse is used to store massive amounts of detailed access data and trajectory data to support efficient analysis and querying, Redis 7 is used to cache hotspot list data to accelerate real-time matching, and Elasticsearch 8 is used to store trajectory indexes to support fast retrieval. The system is deployed in the private cloud environment of the operations and maintenance service center, using Docker containerization technology to achieve independent deployment and elastic scaling of modules, and Prometheus and Grafana to monitor and alert on system operation status.

[0060] The system's front-end provides functional interfaces such as data dashboards, vehicle files, trajectory queries, audit analysis, whitelist management, and system management. The data dashboard displays the percentage and trend data of data processing, image analysis, trajectory reconstruction, abnormal behavior analysis, and manual auditing using pie charts, line charts, and bar charts. The vehicle file page displays the file information and historical trajectories of all vehicles in the database. The trajectory query page supports searching vehicle trajectories by license plate, time, location, etc., and supports map animation to display trajectory details, allowing for the overlay and comparison of multiple trajectories. The audit analysis page displays the trajectories of vehicles identified as abnormal by the abnormal behavior identification module, and supports audit operations on individual trajectories. The whitelist management page supports whitelist maintenance for special vehicles such as rescue vehicles and construction vehicles; the system lowers the alarm priority for whitelisted vehicles during abnormal identification. The system management page provides basic permission management functions such as user, role, department, and road segment management. In summary, this invention constructs an end-to-end data storage and processing architecture, from ETC gantries and toll station data collection equipment to the provincial data aggregation center, and then via fiber optic leased lines and front-end firewalls to a private cloud data platform. It utilizes Kafka message queues to distribute data to different topics according to data type and combines TCP communication protocols and JSON message protocols for real-time push. Simultaneously, a primary / backup redundant link structure ensures the continuity of data transmission. This effectively solves the problems of isolated and scattered data across systems, inability to aggregate cross-road segment data, and the easy backlog and loss of massive high-concurrency data during the collection and transmission stages in existing technologies. It enables traffic data from different road segments and collection points to be uniformly aggregated to the private cloud data platform, providing a foundation for subsequent nationwide vehicle tracking. Trajectory reconstruction provides a complete and reliable data foundation. It resolves toll stations and ETC gantries as nodes in a graph and the reachable paths between any two ordered nodes as edges. Then, in step S6, standardized traffic data is input into the road network model and combined with traffic medium information and license plate recognition information to reconstruct the trajectory. This eliminates reliance on a single data source for vehicle trajectory reconstruction. Even in cases of license plate forgery, obstruction, or recognition failure, the vehicle's location can still be inferred and completed using traffic medium information and road network topology constraints. This overcomes the limitation of traditional license plate recognition, which cannot effectively track vehicles in abnormal situations, achieving accurate reconstruction of the vehicle's true trajectory. It then calls upon pre-configured algorithms for identifying various toll evasion behaviors to analyze the reconstructed vehicle trajectory. The system analyzes toll evasion patterns and integrates the identification logic for various toll evasion behaviors, such as inconsistent vehicle identification tags, masked media, large vehicles with small tags, media swapping, and underpayment, into a unified algorithm model system. This enables the system to automatically match corresponding detection rules and mark the anomaly type for different types of abnormal driving patterns. It overcomes the shortcomings of existing technologies that lack a systematic toll evasion identification model and rely on manual comparison of transaction data, resulting in low screening efficiency and limited coverage. The system achieves automated and systematic identification of diverse toll evasion behaviors and simultaneously generates a complete list of suspected vehicles and evidence chain data. The private cloud data platform encrypts and stores toll identification IDs, license plate numbers, toll fees, and rates using the national cryptographic standard SM2, and combines this with access control policies from a front-end firewall to monitor toll collection. Data transmission between the toll collection network and the operation and maintenance service center is securely isolated, ensuring that sensitive audit data remains encrypted throughout the entire storage chain. Even if the storage medium is illegally accessed, plaintext information cannot be directly obtained, effectively preventing the risk of data leakage during storage and processing. This safeguards the security and compliance of critical data in toll collection audit operations. The list of suspected vehicles and evidence chain data are pushed to the audit front end for manual review. Combined with the vehicle control module, vehicles on the control list are monitored in real time, and the latest location information is pushed to audit personnel when vehicles pass through ETC gantries. Simultaneously, the toll payment work order management module generates toll payment work orders for confirmed toll evasion vehicles, pushes them to the toll payment work order system, and receives review results to synchronously update the blacklist status.This system establishes a closed-loop business process, encompassing anomaly identification, manual review, real-time deployment, toll collection, and blacklist synchronization. It transforms the audit process from reactive post-event tracing to proactive in-event intervention, improving the timeliness and success rate of toll evasion recovery. The system employs a fiber optic leased line access scheme with a primary / backup redundant structure, combined with a two-tiered security isolation mechanism of a front-end firewall and a front-end forwarding server. This allows data transmission between the toll collection network and the operation and maintenance service center to seamlessly switch to the backup link in the event of a primary link failure. Simultaneously, the firewall's access control policies impose fine-grained restrictions on source addresses, destination addresses, and service ports. This ensures both the accessibility of the audit system for vehicle image analysis and evidence collection at toll collection network nodes and achieves secure isolation of the toll collection network. It resolves the issues in existing technologies where frequent connections to the gantry server for data queries can disrupt the billing process and restrict cross-segment trajectory queries, thus guaranteeing the independent and stable operation of the toll collection business system and the audit system.

[0061] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for storing and processing massive audit data based on a private cloud, characterized by: Includes the following steps: S1: Real-time collection of basic vehicle traffic data is achieved through data collection equipment deployed at ETC gantries and toll stations on highways. The basic traffic data includes ETC gantry identification data, ETC gantry transaction data, cash transaction data at toll station entrances, non-cash transaction data at toll station entrances, cash transaction data at toll station exits, and non-cash transaction data at toll station exits. S2: The basic traffic data collected in step S1 is aggregated to the provincial data aggregation center through the toll collection network. The provincial data aggregation center then distributes the data to different topics according to data type through the Kafka message queue, and pushes it in real time using TCP as the communication protocol and JSON as the message protocol. S3: The operation and maintenance service center connects to the provincial data aggregation center's toll network through a leased fiber optic line, receives data pushed by the Kafka message queue through a front-end firewall and a front-end forwarding server, and stores the data in a private cloud data platform. The private cloud data platform adopts a primary and backup link redundancy structure. S4: Standardize the data stored in step S3, unify the data format and encoding, and generate standardized common data; S5: Construct a weighted directed graph-based road network model based on highway network parameters. The road network parameters include toll road information, toll section information, toll unit information, toll station information, toll gantry information, road network node coding information, and adjacent node relationship information. Highway toll stations and ETC gantries are parsed as nodes of the graph, and reachable paths between any two ordered nodes are parsed as edges of the graph. S6: Input the standardized traffic data from step S4 into the road network model from step S5, and combine it with vehicle traffic medium information and license plate recognition information to reconstruct the vehicle's driving trajectory and generate the complete vehicle driving trajectory. S7: Based on the vehicle driving trajectory restored in step S6, call the pre-configured multiple toll evasion behavior recognition algorithm models to analyze the trajectory, identify abnormal behaviors and mark abnormal types, and generate a list of suspect vehicles and corresponding evidence chain data. S8: Push the list of suspected vehicles and evidence chain data generated in step S7 to the audit business front desk for auditors to manually review and confirm, and update the vehicle status based on the review results.

2. The method for storing and processing massive audit data based on a private cloud according to claim 1, characterized in that: The step S2, which distributes data to different topics according to data type, includes: distributing ETC gantry identification data to the first topic, distributing ETC gantry transaction data to the second topic, distributing toll station exit cash transaction data to the third topic, distributing toll station exit non-cash transaction data to the fourth topic, distributing toll station entrance cash transaction data to the fifth topic, and distributing toll station entrance non-cash transaction data to the sixth topic.

3. The method for storing and processing massive audit data based on a private cloud as described in claim 1, characterized in that: The multiple toll evasion behavior identification algorithm models in step S7 include at least the vehicle sign mismatch and OBU misuse identification model, U-turn identification model, abnormal route identification model, shielded medium identification model, one vehicle with multiple media identification model, large vehicle with small sign identification model, medium swapping identification model, and underpayment identification model.

4. The method for storing and processing massive audit data based on a private cloud according to claim 3, characterized in that: The discrepancy in vehicle identification tags and the application of the OBU recognition model identify anomalies through the following steps: 1) Compare the license plates collected from the toll station entry and exit data with the license plates in the toll medium, and cross-compare the license plates identified in the ETC gantry license plate recognition data with the license plates of the toll vehicles in the ETC gantry transaction data. Filter out vehicle information with license plates but no payment and vehicles with payment but no license plates and add it to the pending list. 2) Aggregate the complete passage trajectory of the vehicle by the passage identification ID. When the license plate data and transaction data do not match on 5 or more consecutive ETC gantries, and the similarity between the license plate trajectory and the transaction trajectory is higher than 90%, mark the vehicle trajectory as vehicle license plate mismatch / using OBU and output suspected real license plate.

5. The method for storing and processing massive audit data based on a private cloud according to claim 1, characterized in that: The shielding medium identification model identifies anomalies through the following steps: 1) Based on the passage identifier ID in the ETC gantry transaction data, associate all ETC gantries for the vehicle's passage to form a list of trajectory points, and filter out the trajectory nodes in the ETC gantry transaction data where the license plate of the tolling vehicle is missing and add them to the list to be processed; 2) Based on the start and end times of the billing transaction in the ETC gantry transaction data of this passage, the node range is narrowed down, and the license plate recognition data of the vehicle in the ETC gantry within this range is automatically compared to form a complete path. When the passage meets the condition that there is license plate recognition data but no transaction data in 5 consecutive ETC gantries, the vehicle trajectory is marked as a shielding medium.

6. The method for storing and processing massive audit data based on a private cloud according to claim 1, characterized in that: The large vehicle small label recognition model identifies anomalies through the following steps: 1) Vehicle information is synchronized from the provincial vehicle information database to this system, or the highway owner enters or modifies vehicle information in this system; 2) Match the license plate number and color based on the license plate in the toll station entry and exit data and the license plate of the tolling vehicle in the ETC gantry transaction data. When the vehicle type and number of axles in the toll station entry and exit data, the number of axles and the entry vehicle type in the ETC gantry transaction data are inconsistent with the number of axles and the vehicle type in the vehicle information database of this system, mark the vehicle trajectory as a large vehicle with a small label.

7. The method for storing and processing massive audit data based on a private cloud according to claim 1, characterized in that: The underpayment identification model calculates the amount of evaded fees through the following steps: 1) Using the abnormal driving trajectory reconstructed by the various toll evasion behavior identification algorithm models, combined with the ETC gantry toll rate table information, the theoretical toll is calculated segment by segment according to the gantry sequence through which the trajectory passes; 2) Compare the theoretical toll with the actual amount collected in the toll station exit data. If the actual amount collected is less than the theoretical toll, it is marked as underpayment. The difference is calculated as suspected loss toll and an audit evidence chain including trajectory path, toll calculation details and exit transaction records is generated.

8. The method and system for storing and processing massive audit data based on a private cloud as described in claim 1, characterized in that: In step S3, the private cloud data platform encrypts and stores the vehicle's access ID, license plate number, toll fee, and rate data using the national cryptographic standard SM2.

9. A massive audit data storage and processing system based on a private cloud, used to implement the method as described in any one of claims 1-8, characterized in that: include: The data acquisition module is deployed at highway ETC gantries and toll stations to collect basic vehicle traffic data in real time. The basic traffic data includes ETC gantry identification data, ETC gantry transaction data, toll station entrance cash transaction data, toll station entrance non-cash transaction data, toll station exit cash transaction data, and toll station exit non-cash transaction data. The data aggregation and distribution module is deployed in the provincial data aggregation center. It is used to receive the basic traffic data collected by the data collection module through the toll collection network, and distribute it to different topics according to data type through the Kafka message queue. It uses TCP as the communication protocol and JSON as the message protocol for real-time push. The data receiving and storage module, deployed in the operation and maintenance service center, includes a fiber optic leased line access unit, a front-end firewall unit, a front-end forwarding server unit, and a private cloud storage unit. It is used to access the toll network of the provincial data aggregation center through a leased fiber optic leased line, receive data pushed by the Kafka message queue through the front-end firewall and the front-end forwarding server, and store the data in the private cloud data platform. The private cloud data platform adopts a primary and backup link redundancy structure. The data standardization module is used to standardize the data stored in the data receiving and storage module, unify the data format and encoding, and generate standardized common data. The road network model construction module is used to construct a weighted directed graph form of road network model based on highway road network parameters. The road network parameters include toll road information, toll section information, toll unit information, toll station information, toll gantry information, road network node coding information, and adjacent node relationship information. The module resolves highway toll stations and ETC gantries as nodes of the graph and resolves reachable paths between pairs of ordered nodes as edges of the graph. The trajectory restoration module is used to input the standardized traffic data generated by the data standardization module into the road network model constructed by the road network model construction module, and combine the vehicle traffic medium information and license plate recognition information to restore the vehicle's driving trajectory and generate the complete vehicle driving trajectory. The abnormal behavior identification module is used to analyze the trajectory based on the vehicle driving trajectory restored by the trajectory restoration module, call a variety of pre-configured toll evasion behavior identification algorithm models, identify abnormal behaviors and mark abnormal types, and generate a list of suspect vehicles and corresponding evidence chain data. The audit business processing module is used to push the list of suspected vehicles and evidence chain data generated by the abnormal behavior identification module to the audit business front end for auditors to manually review and confirm, and update the vehicle status based on the review results.

10. The massive audit data storage and processing system based on private cloud according to claim 9, characterized in that: Also includes: The vehicle control module is used to monitor vehicles on the control list in real time. When a vehicle on the control list passes through the ETC gantry, it pushes the latest location information of the vehicle to the auditors. The toll collection work order management module is used to generate toll collection work orders based on manually verified vehicles that have evaded tolls, push the toll collection work orders to the toll collection work order system, receive the review results returned by the toll collection work order system, and update the vehicle blacklist status synchronously. The private cloud storage unit is also used to store vehicle access identification ID, license plate number, toll fee and rate data using the national cryptographic standard SM2.