OBU (On Board Unit) state monitoring method for highway ETC (Electronic Toll Collection) system

Abstract

The invention requests to protect an OBU state monitoring method for an expressway ETC system. The method comprises the following steps: obtaining transaction data and portal topological data of the expressway ETC system; cleaning transaction data and constructing a vehicle track set; comparing the vehicle trajectory set with the portal topology data to identify a missing situation; dividing OBU state levels according to the missing number range; constructing a heterogeneous graph based on the transaction condition of the OBU and the ETC door frame and the topological relation between the ETC door frames; according to the method, a GraphSAGE model is used for classifying OBU nodes in a heterogeneous graph so as to monitor the OBU state, the GraphSAGE model is used for sampling and aggregating neighbor node features, a representation vector of each OBU node is generated, and then real-time monitoring of the OBU state is achieved.

Description

Technical Field

[0001] The invention relates to a highway intelligent transportation system, and in particular to an OBU state monitoring method for a highway ETC system. Background Art

[0002] Digital transformation is profoundly affecting the transportation industry. Promoting the deep integration of new-generation information technologies such as artificial intelligence and the Internet of Things with transportation systems has become a key task in realizing the strategy of building a strong transportation country. In September 2023, the Ministry of Transport clearly pointed out that a comprehensive road network intelligent perception system should be built by comprehensively utilizing technical means such as ETC gantries, communication base stations, video surveillance, radar detection, and meteorological monitoring. Highway ETC data is regarded as a very valuable information resource because of its large data volume, strong temporal nature, and high degree of structuring. These data have shown great application potential in many fields such as traffic flow analysis, congestion prediction, precise navigation, vehicle-road collaboration, and autonomous driving.

[0003] However, the quality of these data is not always consistent, and there are problems such as missing information, which directly affects the effectiveness of subsequent data mining and related research. At present, there are many challenges in real-time monitoring of OBU status. First, some OBU equipment faces outdated technology, degraded performance, and fragility, which makes it difficult to adapt to the needs of modern traffic management. Secondly, the maintenance cost of aging equipment is high, which brings a heavy economic burden to maintenance work. In addition, the equipment lacks the necessary environmental adaptability and is easily affected by natural conditions such as bad weather and extreme temperatures, resulting in failures. At the same time, OBU equipment may also have abnormal charges due to improper human operation or other factors, causing economic losses to relevant departments.

[0004] Through real-time monitoring of the OBU status, comprehensive equipment operation information can be provided to operation and maintenance personnel, maintenance plans can be optimized, and decision support for toll evasion warning can be improved. In addition, the stable operation of OBU equipment will ensure the integrity and validity of data, and provide a solid foundation for scientific research based on big data. Existing equipment monitoring methods mainly conduct single detection in the field of industrial engineering, and do not monitor system equipment at different levels from a macro perspective. Inspired by this, a study on monitoring the OBU equipment status of highway ETC systems was proposed for the first time. Summary of the invention

[0005] The present invention aims to solve the above problems of the prior art. A method for monitoring the OBU status of an ETC system for a highway is proposed. The technical solution of the present invention is as follows:

[0006] A method for monitoring the OBU status of an expressway ETC system, comprising the following steps:

[0007] Step 1, obtaining transaction data and gantry topology data of the highway ETC system;

[0008] Step 2: clean the acquired ETC transaction data, and construct the cleaned data into a vehicle trajectory set in chronological order, where the vehicle trajectory set is defined as a set of multiple ETC gantries that all vehicles pass through during the highway driving process;

[0009] Step 3, comparing the vehicle trajectory set with the ETC gantry topology data to identify missing conditions in the vehicle trajectory set;

[0010] Step 4, classifying the OBU status according to the range of the number of continuity loss;

[0011] Step 5, both OBU and gantry are represented as nodes, and a heterogeneous graph is constructed based on the transaction status of OBU and ETC gantry and the topological relationship between ETC gantry, where OBU as the central node is the main target of classification analysis;

[0012] Step 6: Use the GraphSAGE model to classify the nodes in the heterogeneous graph to monitor the status of the OBU.

[0013] Furthermore, in step 3, the missing conditions in the vehicle trajectory set are identified, and the specific steps are as follows:

[0014] Step 301, initialize an empty list for storing matching results;

[0015] Step 302, comparing each vehicle trajectory set with the ETC gantry topology data node by node;

[0016] Step 303, if the vehicle trajectory set has been traversed but there are still remaining nodes in the transaction sequence, these nodes are identified as missing nodes;

[0017] Step 304: if the current vehicle trajectory set matches the ETC gantry topology data, the node is marked as 0 and moves to the next node of the vehicle trajectory set; if not, the current node is identified as a missing node and continues to match the next node;

[0018] Step 305: Return the matching result list.

[0019] Furthermore, in step 4, the OBU status levels are divided as follows:

[0020] Normal operation state: k = 0;

[0021] Suspicious operating status: k=1;

[0022] Level 1 warning status: 1 <k≤3;

[0023] Level 2 warning status: 4≤k≤6;

[0024] Fault status: 7≤k.

[0025] Furthermore, the specific steps of constructing a heterogeneous graph in step 5 are as follows:

[0026] Step 501, create an empty heterogeneous graph object;

[0027] Step 502, traverse the ETC transaction data, create and add OBU nodes to the heterogeneous graph, and set the corresponding node type;

[0028] Step 503, traverse the ETC gantry topology data, create and add the ETC gantry node to the heterogeneous graph, and set the corresponding node type;

[0029] Step 504, creating edges between the OBU and the ETC gantries and edges between the ETC gantries, adding them to the heterogeneous graph, and setting corresponding edge types;

[0030] Step 505: Set the node type set and the edge type set to the heterogeneous graph object, and output the constructed heterogeneous graph.

[0031] Furthermore, in step 6, the GraphSAGE model is used to learn the representation vector of each OBU node through hierarchical neighbor node sampling and aggregation of neighbor node features, and the OBU nodes in each time window are classified to achieve real-time monitoring of the OBU status.

[0032] The advantages and beneficial effects of the present invention are as follows:

[0033] The present invention automatically acquires and cleans ETC transaction and gantry topology data, constructs a vehicle trajectory set, and compares it with the standard topology to find data missing. Furthermore, the method divides the OBU status into normal, suspicious, first-level warning, second-level warning and fault level according to the missing situation. Using the GraphSAGE algorithm, the present invention constructs a heterogeneous graph, with OBU and gantry as nodes, transactions and topological relationships as edges, and monitors and classifies OBU status in real time by sampling and aggregating neighbor node features. This method has the advantages of high accuracy, real-time performance, easy expansion, reduced economic losses, improved system stability, support for data-driven decision-making, improved user experience, strong environmental adaptability and technological advancement, thereby significantly improving the operating efficiency and reliability of the highway ETC system, and providing operators with a powerful tool for optimizing maintenance strategies and improving service quality.

[0034] The innovation of the present invention is as follows: In step 3, by comparing the vehicle trajectory set with the gantry topology data, data missing is identified. This step is the basis for achieving real-time monitoring and early warning. In step 4, a method for classifying OBU status based on the number of missing data is proposed, which provides a new perspective for the refined management of equipment status. In step 6, the GraphSAGE model is used to classify the OBU nodes in the heterogeneous graph, realizing real-time monitoring of the OBU status. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 The preferred embodiment of the present invention is a schematic diagram of the gantry topology;

[0036] Figure 2 A schematic diagram of a flow chart of an OBU status monitoring method for an expressway ETC system according to the present invention;

[0037] Figure 3 This is a schematic diagram of the percentage of missed transactions;

[0038] Figure 4 This is a flow chart of the GraphSAGE algorithm;

[0039] Figure 5 Results of evaluation indicators under different parameters;

[0040] Figure 6 This is a graph of model performance training results;

[0041] Figure 7 It is the ROC curve diagram;

[0042] Figure 8 The performance indicator diagram of the model under different data amounts;

[0043] Fig. 9 Model prediction time graph for different data amounts;

[0044] Fig.10 This is a graph of the experimental results of the comparison model;

[0045] Fig.11 This is a comparison chart of model testing time. DETAILED DESCRIPTION

[0046] The following will describe the technical solutions in the embodiments of the present invention in detail in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention.

[0047] The technical solution of the present invention to solve the above technical problems is:

[0048] like Figures 1 to 11As shown in one, the present invention discloses an OBU status monitoring method for a highway ETC system, which comprises the following steps:

[0049] Step 1, obtaining transaction data and gantry topology data of the highway ETC system;

[0050] Step 2: clean the acquired ETC transaction data, and construct the cleaned data into a vehicle trajectory set in chronological order, where the vehicle trajectory set is defined as a set of multiple ETC gantries that all vehicles pass through during the highway driving process;

[0051] Step 3, comparing the vehicle trajectory set with the ETC gantry topology data to identify missing conditions in the vehicle trajectory set;

[0052] Step 4, classifying the OBU status according to the range of the number of continuity loss;

[0053] Step 5, both OBU and gantry are represented as nodes, and a heterogeneous graph is constructed based on the transaction status of OBU and ETC gantry and the topological relationship between ETC gantry, where OBU as the central node is the main target of classification analysis;

[0054] Step 6: Use the GraphSAGE model to classify the nodes in the heterogeneous graph to monitor the status of the OBU.

[0055] Furthermore, in step 3, the missing conditions in the vehicle trajectory set are identified, and the specific steps are as follows:

[0056] Step 301, initialize an empty list for storing matching results;

[0057] Step 302, comparing each vehicle trajectory set with the ETC gantry topology data node by node;

[0058] Step 303, if the vehicle trajectory set has been traversed but there are still remaining nodes in the transaction sequence, these nodes are identified as missing nodes;

[0059] Step 304: if the current vehicle trajectory set matches the ETC gantry topology data, the node is marked as 0 and moves to the next node of the vehicle trajectory set; if not, the current node is identified as a missing node and continues to match the next node;

[0060] Step 305: Return the matching result list.

[0061] Furthermore, in step 4, the OBU status levels are divided as follows:

[0062] Normal operation state: k = 0;

[0063] Suspicious operating status: k=1;

[0064] Level 1 warning status: 1 <k≤3;

[0065] Level 2 warning status: 4≤k≤6;

[0066] Fault status: 7≤k.

[0067] Furthermore, the specific steps for constructing a heterogeneous graph in step 5 are as follows:

[0068] Step 501, create an empty heterogeneous graph object;

[0069] Step 502, traverse the ETC transaction data, create and add OBU nodes to the heterogeneous graph, and set the corresponding node type;

[0070] Step 503, traverse the ETC gantry topology data, create and add the ETC gantry node to the heterogeneous graph, and set the corresponding node type;

[0071] Step 504, creating edges between the OBU and the ETC gantries and edges between the ETC gantries, adding them to the heterogeneous graph, and setting corresponding edge types;

[0072] Step 505: Set the node type set and the edge type set to the heterogeneous graph object, and output the constructed heterogeneous graph.

[0073] Furthermore, in step 6, the GraphSAGE model is used to learn the representation vector of each OBU node through hierarchical neighbor node sampling and aggregation of neighbor node features, and the OBU nodes in each time window are classified to achieve real-time monitoring of the OBU status.

[0074] The specific principle of the present invention is described in detail below:

[0075] Definition 1: Standard topology: All gantries can be connected to other gantries through one or more paths, thus forming a connected network and allowing vehicles to pass through.

[0076] Definition 2: Transaction trajectory sequence: refers to the ordered gantry sequence that generates transactions with the vehicle.

[0077] TradTraj= <Flag1,Flag2,…,Flag n >

[0078] Flag n Indicates the mast number

[0079] Definition 3: Missing transactions: Figure 1 (a) is the local standard topology, Figure 1 (b) Figure 1 (c) Figure 1 (d) are three sets of standard topological sequences that can be extracted from the standard topology, where: Figure 1 The two standard topological sequences of (c) are:<K,H,G,E> , indicating that the OBU of the vehicle and the gantry F have missed transactions, such as Figure 1 (e) as shown.

[0080] Definition 4: Missed transaction rate:

[0081] R L is the missed transaction rate, L is the number of missed transactions of OBU, and Trd is the total number of OBU transactions.

[0082] Definition 5: Heterogeneous graph: Graph G = (V, E, T), V = {v1, v2, v3, …, v n}, E={e1,e2,e3,…,e n}, where each node v and each edge e are respectively associated with their mapping function φ(v): V→T V and In addition, T V and T E Represents a collection of object and relationship types, where |T V |+|T E |>2, that is, the node type + edge type in the graph > 2.

[0083] First, based on the ETC original data set, it is pointed out that data missing and abnormal conditions can effectively reflect the operating status of OBU equipment. Then, standard topology is used to conduct an in-depth analysis of data missing conditions, and based on the analysis results, a set of grade standards for the operating status of OBU equipment is formulated. Furthermore, based on the aforementioned equipment status grade standards, a real-time monitoring method for OBU equipment status based on GraphSAGE is proposed. The GraphSAGE model is used to classify heterogeneous graph nodes to monitor the OBU equipment status in real time. The specific flow chart is as follows: Figure 2 shown.

[0084] Before analysis, it is necessary to clean the data, removing redundant and incorrect data to ensure the accuracy and reliability of the data. After cleaning the data, a traditional matching algorithm is used to analyze the transaction track sequence of the transaction data and match it with the standard valid topology to identify the missing situations in the transaction data. First, initialize an empty list `matchedSequences` to store the matching results. Then, for each transaction sequence, perform the matching node by node. If the standard sequence has been traversed but there are still remaining nodes in the transaction sequence, these nodes are identified as missing nodes. If the current node matches the standard sequence, it is marked as 0 and move to the next node of the standard sequence. If it does not match, the current node is identified as a missing node and continue to match the next node. Finally, return the list of matching results. Through the above process, the missing situations in the transaction data are obtained, providing an important basis for the subsequent classification of the status levels of gantry devices and OBU devices.

[0085] After data screening and data cleaning, select the standard topologies with larger trading volumes for subsequent analysis. Match the OBU transaction track set with the standard topologies and analyze the matching result set as Figure 3 shown. It is calculated that the missed transactions of most OBUs are between 0 - 20%, but there will also be OBUs with a higher failure rate.

[0086] Therefore, further analyze the OBUs with a high transaction failure rate and find that the transaction failures are mainly manifested in two situations: occasional and continuous. Conduct a detailed statistical analysis on the OBUs with continuous missing and find that the transaction failures of most OBUs are occasional, but there are also some serious continuous missing, which is very likely the behavior of vehicles deliberately evading tolls. To strengthen the supervision of running vehicles, we define the status levels of OBU devices, where k represents the situation of continuous missing.

[0087] (1) Normal operation status: It means that the OBU generates continuous transaction data within a specific time (k = 0) and there will be no data missing.

[0088] (2) Suspicious operation status: It means that the OBU has occasional transaction failures (k = 1), and it is not yet clear whether this situation is an abnormality caused by the OBU device, only for prompting.

[0089] (3) First-level warning status: It means that the OBU has mild continuous anomalies (1 < k ≤ 3), and the device may have some minor problems or anomalies, and continuous monitoring is required, but it has not yet affected the normal function of the device.

[0090] (4) Intermediate warning status: It means that the OBU has moderate continuous anomalies (4 ≤ k ≤ 6), and there are problems with the device, and some interventions can be carried out and continuous monitoring is continued.

[0091] (5) Fault status: refers to the serious continuous abnormality of the OBU (7≤k), the equipment has serious problems, and has affected normal traffic. The staff needs to check as soon as possible to see whether there is deliberate evasion of fees or other situations.

[0092] The nodes are divided into two categories: OBU nodes and gantry nodes, and the transaction relationship between OBU and gantry (O_G) and the topological relationship between gantries (G_G) are taken as two types of edges.

[0093] First, an empty heterogeneous graph object is created. Then, the transaction data is traversed, OBU nodes are created and added to the heterogeneous graph, and properties are set for each node. At the same time, the node type "OBU" is added to the node type collection. Next, the topology data is traversed, gantry nodes are created and added to the heterogeneous graph, and properties are set for each node. While creating the nodes, two types of edges are also created and added to the heterogeneous graph. At the same time, edge types O_G and G_G are added to the edge type collection. Finally, the node type collection and edge type collection are set to the heterogeneous graph object, and the heterogeneous graph is output.

[0094] GraphSAGE is a graph neural network model that can effectively capture the local structural information of nodes. By sampling neighbor nodes and aggregating their features, it can reveal the connections and features between nodes. It is suitable for processing large-scale graph data, and can solve the problems of computational and memory limitations by approximately aggregating neighbor nodes. In addition, GraphSAGE combines the attribute features of nodes with the topological structure of the graph by learning the embedded representation of nodes. It can capture the position of nodes in the graph and the information of surrounding nodes, thereby improving the accuracy of node classification tasks. GraphSAGE can also be applied to time window modeling. By learning the embedded representation of nodes and the state scoring function, it can dynamically monitor the state changes of nodes in the time window. Figure 4 The specific process of the algorithm is shown in FIG.

[0095] In order to find out the impact of the total number of layers L and hidden states d in GraphSage on the overall model. The experimental learning rate is set to 0.01 and the weight decay is set to 5×10 -4 The maximum number of training times is 200. The range of L is {1, 2, 3}, the range of d is {4, 8, 16, 32, 64}, and other parameters are set to the optimal value. Figure 5As shown in the four figures, we can see the impact of different L and d values ​​on accuracy, recall, precision and F1-Score. Overall, the color distribution shows the trend of the evaluation index changing with the change of L and d values. When d = 4, the value of the evaluation index first decreases and then increases with the increase of L value. Further observation shows that the value of the evaluation index continues to increase with the increase of d and L values. In particular, when d = 8 and L = 3, all evaluation indicators reach the optimal value. In addition, when L = 3, the values ​​of each evaluation indicator tend to be stable, indicating that d and L have reached the optimal configuration.

[0096] During full batch training of the model, Figure 6 The training performance of the model is shown. The blue line tracks the change of loss value under different epochs: in the first 50 epochs, the loss value drops rapidly. As the epoch increases, the rate of decrease of the loss value gradually slows down and stabilizes in the later stage. At the same time, the figure also shows four other lines, representing different evaluation indicators of the model under different epochs. In the range of 0 to 50 epochs, the values ​​of these evaluation indicators rise rapidly, accompanied by slight fluctuations. When the epoch value is in the range of 50 to 100, the values ​​of the evaluation indicators continue to rise at a relatively stable rate. After the epoch value exceeds 100, the values ​​of these evaluation indicators begin to stabilize.

[0097] During the model training process, the significant improvement in accuracy reflects the enhancement of model performance. In the first few epochs, the rapid growth of accuracy indicates that the model begins to effectively classify samples. As training continues, although the rate of improvement in accuracy may slow down, it still shows a steady upward trend overall. Precision, recall, and F1 value may fluctuate greatly in the first few epochs, but as training progresses, these indicators gradually stabilize. A higher precision indicates that the model is more accurate in predicting positive examples, while a higher recall indicates that the model is better able to find positive examples.

[0098] The area under the ROC curve is an important model evaluation indicator, which measures the model's ability to distinguish positive and negative examples at different decision thresholds. Figure 7 As can be seen, the AUC values ​​of categories 0, 1, 2, 3, and 4 do not change much and are close to 1, which indicates that the models for these categories have high classification accuracy. Although the ROC curve for category 1 shows some fluctuations, its AUC value is 0.8, which still shows that the model has good classification performance at different thresholds.

[0099] An AUC value close to 1 usually means that the model has a high true positive rate and a low false positive rate, and can maintain a high classification ability even at different decision thresholds. This performance further verifies that the model has good robustness and stability and can provide reliable prediction results under various conditions.

[0100] In order to verify the performance of the model under different data amounts, the changes of various evaluation indicators (Acc, Pre, Rec, F1) of the model under different data amounts are analyzed, such as Figure 8 As shown in the figure, the four figures are the changes of different performance indicators under different numbers of nodes, the horizontal axis is the number of nodes, and the vertical axis is the value of the evaluation indicator. Overall, as the amount of data increases, the values ​​of each indicator fluctuate slightly, with a fluctuation range of 0.002. The evaluation index values ​​of the overall model are all above 99%, indicating that the change in data volume has little effect on the model, proving the effectiveness and stability of the algorithm model in this chapter, and the model has good generalization ability.

[0101] In order to further verify the impact of data size on model performance, the experiment selected four data sets with different numbers of nodes for verification. The results are as follows Fig. 9 As shown in the figure, the horizontal axis represents the number of nodes, and the y-axis represents the model test time. As the amount of data increases, the overall time shows an upward trend. As the amount of data increases, the increase in model test time gradually decreases, indicating that as the amount of data increases, the model test time does not fluctuate much and shows a convergence trend.

[0102] In order to further verify the performance of the proposed model, DeepWalk, GCN, SVM, XGBoost, GAT, MST-GAT and CST-GL are used as comparison models to evaluate the performance of the GraphSAGE model. Fig.10 By observing the chart, we can intuitively find that the proposed model shows the best performance in all evaluation indicators. In the table, we can clearly see that the evaluation index values ​​are all higher than 99%. These results further verify the superiority of the proposed model in performance.

[0103] Without considering the graph structure information, the performance of XGBoost and SVM is about 97%. SVM is known for its ability to handle high-dimensional feature spaces and construct nonlinear decision boundaries, while XGBoost shows its advantages in handling complex nonlinear relationships, high-dimensional features, and large-scale data sets. However, compared with the GraphSAGE model, a major limitation of SVM is that it cannot directly utilize the information of graph structure and neighbor nodes, which is crucial for the analysis of graph data. Similarly, although XGBoost performs well in many machine learning tasks, it does not explicitly consider the information of graph structure and neighbor nodes, which may be one of the reasons why its performance is not as good as GraphSAGE.

[0104] In the general graph model, GCN is a graph neural network model based on graph convolution operations, which learns the representation of each node by convolving the features of the node with the features of the neighboring nodes. GAT is a graph neural network model based on the attention mechanism, which aggregates the information of neighboring nodes by learning the attention weights between each node and the neighboring nodes. GAT performs well in processing node classification tasks with complex connection patterns in graphs. However, in this study, the nodes are divided into two categories corresponding to the two layers of nodes in GraphSAGE. This model can better extract the features and connection relationships of the OBU core nodes, so GraphSAGE still achieves better results compared with these two models.

[0105] In addition, DeepWalk is a graph embedding method based on random walks, which captures the relationship between nodes by learning vector representations of nodes. Compared with GraphSAGE, DeepWalk focuses more on generating low-dimensional embeddings of nodes. DeepWalk performs poorly in node classification tasks because it only learns the static features of nodes without utilizing the neighbor information of nodes.

[0106] MST-GAT uses multimodal graph attention networks and temporal convolutional networks to capture spatial correlation and temporal dependency between multimodal time series. Compared with GraphSAGE, MST-GAT pays too much attention to spatial correlation and has a slightly lower accuracy in classifying OBU nodes. CST-GL can automatically detect anomalies in complex time series through automatic thresholds and effectively perform early detection, but it captures the pairwise interdependencies between time series variables and is slightly less accurate in classifying highway OBU status than GraphSAGE.

[0107] according to Fig.11The prediction time analysis of different models shown in the figure does not include the real-time data processing time, but only considers the test time of the data in the model. It can be observed that the model proposed in this chapter takes the shortest time to predict, while the GAT model takes the longest time. Although the prediction accuracy of the GAT model is similar to that of the model proposed in this chapter, it takes longer. Comprehensively analyzing the prediction accuracy and time efficiency of each model, the model proposed in this chapter achieves the best in both prediction accuracy and time performance.

[0108] Regarding the methods in the above technical briefing, we can further expand and enrich their application scenarios and functions in the following ways to improve the stability and efficiency of the system:

[0109] In addition, the present invention can also enrich and expand the above embodiments as follows, which all belong to the protection scope of the present invention.

[0110] 1. Introducing multi-data source fusion analysis

[0111] - Fusion of video surveillance data: By analyzing the video surveillance of vehicles passing through the ETC gantry, it is possible to verify whether the physical status of the OBU equipment is normal, such as whether the license plate recognition is clear and whether the OBU is blocked or displaced.

[0112] -Integrate meteorological data: Consider the potential impact of weather conditions (such as rain, snow, and thunderstorms) on OBU reading. Meteorological data can be used to correct for increased transaction failure rates caused by bad weather.

[0113] - Utilizing radar detection data: Radar data can provide vehicle speed and position information, which helps to identify whether the lack of continuity of the OBU is related to abnormal vehicle speed or driving behavior.

[0114] 2. Enhance the dynamic adjustment mechanism of OBU status level

[0115] -Introduction of the concept of time window: When dynamically monitoring the OBU status, time windows of different lengths can be set to adjust the OBU status level in real time according to recent transaction conditions, which can more accurately reflect the instantaneous status of the OBU.

[0116] - Incorporate user feedback: Allow drivers to report OBU problems through mobile apps or customer service hotlines, incorporate user feedback into status assessments, and improve the comprehensiveness and timeliness of inspections.

[0117] 3. Optimizing heterogeneous graph construction and GraphSAGE model training

[0118] -Graph structure enhancement: In heterogeneous graphs, in addition to OBU and gantry nodes, other types of nodes (such as road section nodes and meteorological nodes) can also be introduced to enhance the expressiveness of the graph and the prediction accuracy of the model.

[0119] - Feature engineering improvements: The features of OBU and gantry nodes are processed more finely, such as adding information such as vehicle type, OBU model, gantry position, etc., so that the GraphSAGE model can capture more correlations.

[0120] -Hyperparameter optimization: Use more intelligent algorithms (such as Bayesian optimization, genetic algorithms, etc.) to automatically find the best hyperparameter combination for the GraphSAGE model to further improve the performance of the model.

[0121] 4. Intelligent early warning and intervention of abnormal trading behavior

[0122] -Behavioral pattern analysis: Combine historical data and use clustering algorithms or time series analysis to identify patterns of abnormal trading behavior, such as frequent short-distance transactions, high transaction failure rates in specific areas, etc.

[0123] -Real-time intervention mechanism: When the OBU status reaches the warning level, the system automatically sends a prompt message to the driver, recommending an OBU inspection or update, and notifies the maintenance team to conduct an on-site inspection.

[0124] 5. AI-based predictive maintenance

[0125] - Predict equipment failure: Use the trained GraphSAGE model to predict the failure probability of the OBU within a certain period of time in the future, and perform maintenance or replacement in advance.

[0126] -Optimize maintenance strategy: Develop the best maintenance plan based on fault prediction, reduce unnecessary maintenance work and improve maintenance efficiency.

[0127] 6. Efficient data management and privacy protection

[0128] -Data tiered storage: Based on the importance of data, a tiered storage strategy is adopted, with key transaction data stored in real-time cache and historical data stored in lower-cost storage devices.

[0129] -Differential privacy: Differential privacy technology is introduced during data processing and analysis to protect users' transaction information from being leaked and enhance users' trust in the system.

[0130] 7. Realize cross-regional OBU status monitoring

[0131] - Standardized data interface: Design a unified data interface standard to enable different highway ETC systems to share OBU status information and improve the feasibility of cross-regional monitoring.

[0132] -Federated Learning: Using federated learning technology, different ETC systems can jointly train the GraphSAGE model without sharing the original data, thereby improving the generalization ability of the model.

[0133] 8. Enhance the model’s explanatory power

[0134] - Feature importance analysis: The contribution of each feature in the GraphSAGE model is analyzed through methods such as Shapley value to help understand the model's decision logic for OBU status classification.

[0135] -Visualization tools: Develop visualization tools to display the location and status changes of OBU nodes in heterogeneous graphs, and improve the system operation and maintenance personnel's understanding of monitoring results.

[0136] 9. Software and hardware collaborative optimization

[0137] - Hardware acceleration: Research the application of GPU or dedicated accelerators (such as TPU) in GraphSAGE model training and inference to improve data processing speed and model prediction efficiency.

[0138] -Software optimization: optimize algorithm implementation, reduce unnecessary computing overhead, and improve the training speed and real-time monitoring performance of the GraphSAGE model.

[0139] 10. User experience and service improvement

[0140] -Convenient OBU update service: Develop a mobile APP that allows users to update the OBU software online and reduce transaction anomalies caused by outdated OBU software versions.

[0141] -Personalized notifications and suggestions: Based on the OBU status monitoring results, users are provided with personalized maintenance reminders and travel suggestions, such as suggesting to replace OBU equipment, avoid ETC gantries with high failure rates, etc.

[0142] Through the above expansion and enrichment, the OBU status monitoring method for highway ETC systems will be able to more comprehensively and intelligently identify and process the operating status of OBU equipment, effectively prevent and reduce transaction anomalies, and improve the overall performance and service quality of highway ETC systems.

[0143] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0144] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0145] The above embodiments should be understood to be only used to illustrate the present invention and not to limit the protection scope of the present invention. After reading the contents of the present invention, technicians can make various changes or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims

1. A method for monitoring the OBU status of an expressway ETC system, characterized in that: The following steps are involved: Obtain transaction data and gantry topology data of the highway ETC system; clean transaction data and construct vehicle trajectory sets; compare vehicle trajectory sets with gantry topology data to identify missing conditions; classify OBU status levels according to the range of missing quantity; A heterogeneous graph is constructed based on the transaction status of OBU and ETC gantries and the topological relationship between ETC gantries. The OBU nodes in the heterogeneous graph are classified using the GraphSAGE model to monitor the OBU status. The GraphSAGE model generates a representation vector for each OBU node by sampling and aggregating the features of neighbor nodes, thereby realizing real-time monitoring of the OBU status.

2. The OBU status monitoring method according to claim 1, characterized in that: The identifying missing conditions in the vehicle trajectory set specifically includes: initializing a matching result list; comparing the vehicle trajectory set with the gantry topology data node by node; identifying unmatched nodes in the transaction sequence as missing nodes; and returning a matching result list.

3. The OBU status monitoring method according to claim 1, characterized in that: The OBU status level classification includes: dividing the OBU status into normal operating status, suspicious operating status, first-level warning status, second-level warning status and fault status, wherein the normal operating status refers to no data missing; the suspicious operating status refers to occasional transaction failures; the first-level warning status refers to mild continuous abnormalities; the second-level warning status refers to moderate continuous abnormalities; and the fault status refers to severe continuous abnormalities.

4. The OBU status monitoring method according to claim 1, characterized in that: The heterogeneous graph construction step also includes: creating a heterogeneous graph object; setting types for OBU and gantry nodes; adding transaction relationship and topological relationship edges to the heterogeneous graph; and outputting the constructed heterogeneous graph.

5. The OBU status monitoring method according to claim 1, characterized in that: The GraphSAGE model samples neighboring nodes and aggregates features for the OBU nodes in each time window to learn the representation vector of each OBU node to achieve real-time monitoring of the OBU status.

6. The OBU status monitoring method according to claim 1, characterized in that: The method further includes: based on the parameter configuration of the GraphSAGE model, determining that the total number of layers is 2 and the hidden state size is 32 to achieve optimal model performance.

7. The OBU status monitoring method according to claim 1, characterized in that: The method further optimizes the test time of the model under different data amounts, so that as the amount of data increases, the test time of the model shows a convergence trend and the fluctuation range is within a reasonable range.

8. The OBU status monitoring method according to claim 1, characterized in that: The method compared multiple models, including DeepWalk, GCN, SVM, XGBoost, GAT, MST-GAT and CST-GL, and the results showed that the GraphSAGE model performed best in terms of classification accuracy and time efficiency.

9. The OBU status monitoring method according to claim 1, characterized in that: The method takes advantage of heterogeneous graphs and regards OBU equipment and gantry equipment as different types of nodes, thereby enhancing the robustness and accuracy of OBU status monitoring.

Images