Weighing abnormity alarm and whole-process intelligent management and control method and system

Abstract

The invention discloses a weighing abnormity alarm and whole-process intelligent management and control method and system, and relates to the field of logistics transportation monitoring and intelligent identification, and the method comprises the steps: collecting the weighing behavior data of a vehicle, and constructing a vehicle feature vector based on the weighing behavior data; clustering the vehicle feature vectors of all vehicles to obtain a plurality of vehicle clusters with similar behavior patterns; matching the vehicle clusters with the corresponding transportation tasks, and extracting matching features in combination with information of the vehicle clusters and the transportation tasks; inputting the matching features into a trained cluster-level anomaly recognition model, and outputting an anomaly score of each vehicle cluster; and when the abnormal score of a certain vehicle cluster exceeds a preset risk threshold value, weighing abnormal warning information is output to a process control module. According to the invention, vehicle clusters with similar behavior patterns can be mined from an overall perspective, and potential abnormal behaviors of suspected multi-person collaborative crime committing are recognized by using the cluster-level anomaly recognition model, so that the problem of poor overall collaboration of weighing anomaly recognition in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of logistics transportation monitoring and intelligent identification, and more specifically, to a method and system for abnormal weighing alarm and full-process intelligent management and control. Background Art

[0002] In industrial transport scenarios such as waste recycling, building materials transportation, and mineral logistics, weighing (ground scale weighing) is often used as the primary basis for transport billing and settlement. Vehicles are weighed for both tare and gross weight upon entering or leaving the factory. The net weight is calculated from the difference, and the quantity or volume of the transported materials is then confirmed.

[0003] Traditional weighing anomaly identification systems are mostly based on single-vehicle dimension rule settings, such as limiting the frequency of weighing the same license plate within a certain period of time, identifying abnormal weight fluctuations of the same vehicle, or single-vehicle routes that do not match the task, etc. This type of method is more effective when facing repeated crimes committed by a single vehicle and obvious data anomalies. However, with the gradual improvement of the supervision mechanism, the actual method of committing crimes has also tended to be covert and organized, and more covert "group-style" abnormal behaviors have emerged. For example, for medium and large-scale transportation tasks, especially in situations where the demand for a single material is large, it is often necessary for multiple vehicles to complete a transportation task together, that is, a collaborative transportation mode of "one task corresponds to multiple vehicles." Although this multi-vehicle collaboration mode has improved transportation efficiency, it has also brought more complex supervision problems.

[0004] In practice, unscrupulous transport groups exploit this weak regulatory link to implement various forms of coordinated fraud. For example, they split a task that could have been completed by three vehicles into five, leaving each vehicle underloaded. This results in an undercount of the total net weight, which in turn leads to an over-count of transport trips and the resulting excess transportation fees. Some vehicles deliberately extend transport routes by detours or modify GPS track data to defraud surcharges based on distance or time. Some vehicles inject large amounts of water into the interlayer of the vehicle before weighing the empty vehicle, increasing the tare weight before draining the water and loading the cargo. This creates an artificial bias in the net weight calculation and inflates the settlement tonnage. Multiple vehicles enter and exit in batches, staggering weighing times, in an attempt to circumvent behavioral correlation analysis for the same task.

[0005] Because multiple vehicles collaborate on a transport mission, the focus of supervision is no longer on the behavior of individual vehicles, but rather the overall performance of the entire fleet. Traditional weighing monitoring methods typically analyze indicators such as individual vehicle tare weight fluctuations, frequency anomalies, and loading ratios, making it difficult to effectively identify collective abnormal behavior among multiple vehicles. Furthermore, existing systems lack a mechanism for comparing and analyzing transport missions and vehicle execution processes, making it impossible to accurately determine whether a particular transport mission presents a risk of collective fraud in actual operations.

[0006] Therefore, there is an urgent need for an intelligent weighing monitoring solution that can identify the collaborative behavior of multiple vehicles and conduct aggregate analysis from a holistic perspective, thereby achieving early identification and graded warning of collaborative transportation behavior of multiple people. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a weighing abnormality alarm and full-process intelligent management and control method and system to solve the problems mentioned in the background technology.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] A weighing abnormality alarm and full-process intelligent management and control method includes the following steps:

[0010] Collecting vehicle weighing behavior data and constructing a vehicle feature vector based on the weighing behavior data;

[0011] Cluster the vehicle feature vectors of all vehicles to obtain multiple vehicle clusters with similar behavior patterns;

[0012] Matching the vehicle cluster with its corresponding transport task, and extracting matching features based on information of the vehicle cluster and the transport task;

[0013] Input the matching features into the trained cluster-level anomaly recognition model and output an anomaly score for each vehicle cluster;

[0014] When the abnormal score of a vehicle cluster exceeds the preset risk threshold, the weighing abnormality alarm information is output to the process control module.

[0015] In some embodiments, the weighing behavior data includes any one or more of the following: vehicle model, vehicle rated tonnage, weighing time, weighing weight, transportation route, loading ratio, and GPS track.

[0016] In some embodiments, the matching features include any one or more of the following:

[0017] the deviation ratio between the total loaded weight of the vehicle cluster and the nominal total weight of the transport task;

[0018] The average trajectory similarity between the transport trajectory of each vehicle in the vehicle cluster and the preset trajectory of the transport task;

[0019] The average overlap between the actual arrival time of the vehicles in the vehicle cluster and the specified execution time window of the transport task;

[0020] The average proportion of the trajectory paths of the vehicles in the vehicle cluster covering the transportation task path;

[0021] the standard deviation of the loaded weights of the vehicles in the vehicle cluster;

[0022] The average loading rate of the vehicle cluster.

[0023] In some embodiments, the cluster-level anomaly recognition model is a binary classification or regression scoring model trained based on supervised learning, and the model uses multiple historically labeled normal transportation clusters and abnormal transportation clusters as training samples.

[0024] The matching features are input and the output is a probability score of whether each vehicle cluster is abnormal transportation.

[0025] In some embodiments, the training process of the cluster-level anomaly recognition model includes the following steps:

[0026] Construct multiple vehicle clusters based on historical transportation data. Each vehicle cluster is generated by a clustering algorithm and contains vehicle feature vectors of all vehicles in the cluster and their corresponding transportation task information.

[0027] Manually label the vehicle clusters or perform system reconciliation labeling to obtain a label indicating whether each vehicle cluster is an abnormal transport cluster;

[0028] Extract the matching features for each vehicle cluster as the input vector of the training sample, and use the label as the training output;

[0029] The cluster-level anomaly recognition model is trained using a supervised learning method.

[0030] In some embodiments, the clustering method is a density-based clustering algorithm or a hierarchical clustering algorithm based on a similarity threshold.

[0031] In some embodiments, the method further includes: after determining that a certain vehicle cluster is an abnormal cluster, identifying suspected abnormal vehicles from the abnormal cluster, the identification process including:

[0032] Extracting the vehicle feature vector of each vehicle in the abnormal cluster;

[0033] Inputting the vehicle feature vector into an abnormal vehicle identification model and outputting an abnormality score for each vehicle in the abnormal cluster;

[0034] Based on the anomaly score, one or more vehicles in the vehicle cluster having an anomaly score higher than a preset threshold are determined as suspected abnormal vehicles.

[0035] In some embodiments, the training process of the abnormal vehicle identification model includes the following steps:

[0036] A training sample set is constructed based on historical transportation records and transportation task reconciliation information. Each sample corresponds to the vehicle feature vector of a vehicle.

[0037] Based on manual verification results, the vehicle is labeled to identify whether it is an abnormal vehicle;

[0038] A supervised learning algorithm is used to train the abnormal vehicle recognition model.

[0039] In some embodiments, the process control module includes an alarm recording submodule, a linkage management and control submodule, and a manual review submodule;

[0040] The alarm recording submodule is used to receive and record the abnormal weighing alarm information and the IDs of all vehicles in the corresponding vehicle cluster;

[0041] The linkage control submodule is used to trigger a control strategy for vehicles in an abnormal vehicle cluster after receiving abnormal alarm information;

[0042] The manual review submodule is used to push suspected abnormal vehicle clusters to the operation and maintenance platform and provide a visual analysis interface for manual confirmation.

[0043] The present invention also discloses a weighing abnormality alarm and full-process intelligent management and control system, which includes the following modules:

[0044] A vehicle feature construction module, configured to collect vehicle weighing behavior data and construct a vehicle feature vector based on the weighing behavior data;

[0045] The vehicle clustering module is used to cluster the vehicle feature vectors of all vehicles to obtain multiple vehicle clusters with similar behavior patterns;

[0046] A task matching feature extraction module is used to match the vehicle cluster with its corresponding transportation task and extract matching features based on the information of the vehicle cluster and the transportation task;

[0047] A cluster-level anomaly scoring module is used to input the matching features into a trained cluster-level anomaly recognition model and output an anomaly score for each vehicle cluster;

[0048] The abnormal alarm trigger module is used to output the weighing abnormal alarm information to the process control module when the abnormal score of a vehicle cluster exceeds the preset risk threshold.

[0049] The advantage of the present invention over the existing technology is that, compared with the existing weighing monitoring methods that rely on single-vehicle rule detection or manual spot checks, the present invention constructs vehicle feature vectors and introduces clustering algorithms to mine vehicle clusters with similar behavior patterns from the overall transportation behavior, thereby automatically classifying vehicle groups with potential collaborative behavior characteristics; combining the matching features between vehicle clusters and their corresponding transportation tasks, and inputting them into the cluster-level anomaly recognition model, an anomaly score can be generated for each vehicle cluster, thereby identifying transportation behaviors suspected of being coordinated by multiple people, and having the ability to identify collaborative anomalies from a holistic perspective.

[0050] Furthermore, for identified anomalous vehicle clusters, the system extracts features from each vehicle within the cluster and uses the anomalous vehicle identification model to assess its individual anomaly score, thereby identifying suspected problem vehicles down to the individual vehicle level. This collaborative identification of both cluster and individual vehicle models not only improves anomaly identification accuracy but also assists in determining whether the anomalous cluster represents organized group transport or merely individual vehicle violations, facilitating subsequent manual review and targeted remediation.

[0051] In addition, the present invention can be linked with the process control module to automatically record and trigger linkage strategies after generating abnormal alarms, or push suspicious clusters to the manual review platform, thereby improving the intelligence and risk control response efficiency of the weighing process.

[0052] The present invention also has the ability to automatically cluster and categorize vehicles. By modeling vehicle weighing behavior data and extracting vehicle characteristics, the system can automatically identify the most likely transport task for each vehicle in a complex, multi-tasking environment, improving task matching accuracy and management transparency. Leveraging modeled matching feature calculation and task constraint verification mechanisms, it can effectively assist enterprises in establishing automated vehicle-task management relationships, avoiding billing discrepancies, unclear responsibilities, and subsequent evidence collection difficulties caused by task confusion. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is the overall flow chart of the present invention;

[0054] Figure 2 This is a flow chart of the cluster-level anomaly recognition model training process of the present invention;

[0055] Figure 3 This is a flow chart of abnormal vehicle identification according to the present invention;

[0056] Figure 4 This is a training flow chart of the abnormal vehicle recognition model of the present invention. DETAILED DESCRIPTION

[0057] The specific embodiments of the present invention will be described below with reference to the accompanying drawings.

[0058] This method aims to achieve monitoring, early warning and full-process control of abnormal transportation behavior by collecting and analyzing vehicle weighing behavior data and combining it with intelligent recognition technology.

[0059] like Figure 1 Shown is a flow chart of the overall scheme of the present invention, comprising the following steps:

[0060] Collecting vehicle weighing behavior data and constructing a vehicle feature vector based on the weighing behavior data;

[0061] Cluster the vehicle feature vectors of all vehicles to obtain multiple vehicle clusters with similar behavior patterns;

[0062] Matching the vehicle cluster with its corresponding transport task, and extracting matching features based on information of the vehicle cluster and the transport task;

[0063] Input the matching features into the trained cluster-level anomaly recognition model and output an anomaly score for each vehicle cluster;

[0064] When the abnormal score of a vehicle cluster exceeds the preset risk threshold, the weighing abnormality alarm information is output to the process control module.

[0065] In a more specific embodiment, it is first necessary to collect the behavior data of the vehicle during the weighing process. These data are the basis for subsequent analysis, covering various attributes of the vehicle and dynamic information during transportation. Specifically, the collected data may include vehicle model (such as heavy truck, medium truck, etc.), vehicle rated tonnage (that is, the maximum load allowed by the vehicle design, such as 20 tons), weighing time (timestamp accurate to the second, such as 2023-10-1514:35:22), weighing weight (actual weighing result, such as 18.5 tons), transportation route (complete route description from the starting point to the end point), loading ratio (the ratio of actual weight to rated tonnage, such as 92.5%) and GPS trajectory (sequence of latitude and longitude coordinates during vehicle driving). The collection of these data can be completed collaboratively through weighing equipment, GPS positioning system and transportation management system.

[0066] After collecting this data, it needs to be integrated into a multi-dimensional vehicle feature vector to facilitate subsequent mathematical modeling and analysis. For example, a vehicle feature vector can be represented as [vehicle model code, rated tonnage, weighing timestamp, weighing weight, transport route code, loading ratio, GPS trajectory sequence].

[0067] After constructing the feature vectors for all vehicles, the next step is to perform cluster analysis on these vectors to identify groups of vehicles with similar behavior patterns. The core idea of ​​clustering is to group vehicles with similar transportation behaviors into a cluster, while the behaviors of different clusters vary greatly.

[0068] During clustering, a density-based clustering algorithm (such as DBSCAN) or a hierarchical clustering algorithm based on a similarity threshold may be used in some embodiments.

[0069] The advantage of DBSCAN is that it does not require a predefined number of clusters. Key parameters include the neighborhood radius ε (which can be selected from 0.1 to 1 in some embodiments, adjusted based on data density) and the minimum number of points MinPts (which can be selected from 3 to 10 in some embodiments, depending on the total number of vehicles). Hierarchical clustering gradually merges vehicles by calculating the similarity between feature vectors (such as Euclidean distance or cosine similarity) until a preset similarity threshold (which can be selected from 0.7 to 0.9 in some embodiments) is reached.

[0070] For example, suppose we have 100 vehicle feature vectors and use DBSCAN clustering with ε = 0.5 and MinPts = 5. We might obtain five vehicle clusters and several noise points. Vehicles within each cluster exhibit high consistency in weighing time, weight, and route. After clustering, each vehicle is assigned to a cluster, and subsequent analysis focuses on these clusters.

[0071] Next, each vehicle cluster is matched to its corresponding transport mission. Transport missions are typically provided by the management system and include information such as a nominal gross weight (e.g., 100 tons), a predefined trajectory (e.g., "origin A - transfer B - destination C"), and a specified execution time window (e.g., 2023-10-15 14:00 to 16:00). The purpose of this matching is to assess whether the vehicle cluster is performing according to the mission requirements.

[0072] During the matching process, a series of features need to be extracted to quantify the degree of matching. These features include the following aspects:

[0073] 1. The deviation ratio between the total loaded weight and the nominal total weight. This is calculated as: Deviation ratio = |Cluster total loaded weight - Nominal total weight| / Nominal total weight. For example, if the total loaded weight of three vehicles in a cluster is 58 tons and the nominal total weight of the mission is 60 tons, the deviation ratio is |58 - 60| / 60 = 0.0333 (or 3.33%). The smaller the deviation ratio, the more the cluster meets the mission requirements.

[0074] 2. Average trajectory similarity between the transport trajectory and the pre-set trajectory. The Dynamic Time Warping (DTW) algorithm is used to calculate the similarity between each vehicle's GPS trajectory and the pre-set trajectory and take the average. The similarity ranges from 0 to 1, with higher values ​​indicating closer trajectory alignment. For example, if the similarities of three vehicles in a cluster are 0.9, 0.85, and 0.95, respectively, the average similarity is (0.9 + 0.85 + 0.95) / 3 = 0.9.

[0075] 3. The average overlap between the actual arrival time and the specified execution time window. Specifically, for each vehicle, calculate the proportion of its actual arrival time within the time window. For example, if the time window is from 2:00 PM to 4:00 PM, and a vehicle arrives at 2:30 PM, the overlap is 1; if it arrives at 4:30 PM, the overlap is 0. The average overlap within a cluster is the mean of the overlaps for all vehicles.

[0076] 4. The average percentage of the trajectory path covering the transport mission path. Specifically, calculate the percentage of each vehicle's GPS trajectory covering the preset path and take the average. For example, if the mission path is 100 kilometers and a vehicle covers 90 kilometers, the coverage percentage is 0.9.

[0077] 5. Standard deviation between loading weights. Specifically, the standard deviation of the loading weights of each vehicle in the cluster is calculated. The smaller the standard deviation, the higher the loading consistency.

[0078] 6. Average Loading Rate: Specifically, calculate the average loading rate of all vehicles in the cluster. Loading rate is defined as: weighed weight / approved tonnage. For example, if three vehicles have loading rates of 0.9, 0.95, and 0.85, respectively, the average loading rate is (0.9 + 0.95 + 0.85) / 3 = 0.9. A higher loading rate indicates greater space utilization and a lower likelihood of fraud.

[0079] These features will be used as input for subsequent anomaly identification. For example, the matching feature vector of a cluster may be [0.0333, 0.9, 0.95, 0.9, 2.5, 0.9], corresponding to the above 6 features.

[0080] After extracting matching features, they are fed into a trained cluster-level anomaly recognition model, which outputs an anomaly score for each vehicle cluster. The cluster-level anomaly recognition model of the present invention is a binary classification or regression scoring model based on supervised learning. The output value is a probability score between 0 and 1, indicating the likelihood that the cluster is an abnormal transport.

[0081] like Figure 2 As shown in Figure 2, the model training process includes the following steps:

[0082] Use historical transportation data to generate multiple vehicle clusters, each containing a feature vector and corresponding transportation task information. For example, historical data may contain 1000 clusters, of which 800 are normal and 200 are abnormal.

[0083] Each cluster is marked as abnormal through manual inspection or system reconciliation (such as checking shipping documents). Abnormal clusters may be manifested as serious deviations in loading weight, inconsistent trajectory, etc.

[0084] Extract the above matching features for each cluster as the input vector. For example, the feature vector of an abnormal cluster may be [0.5, 0.6, 0.4, 0.5, 10, 0.7].

[0085] Train the model using a supervised learning algorithm (such as logistic regression, support vector machine, or neural network). Assuming a neural network is used, the architecture can include an input layer (6 neurons corresponding to 6 features), 2 hidden layers (32 neurons each, using the ReLU activation function), and an output layer (1 neuron, using the Sigmoid function to output probabilities). Training uses the cross-entropy loss function and the Adam optimizer. In some embodiments, the learning rate can be 0.001, and the number of training rounds is 100.

[0086] In the application, suppose the matching features of a cluster are [0.0333, 0.9, 0.95, 0.9, 2.5, 0.9], and the model outputs an anomaly score of 0.1, indicating a low possibility of anomaly; while the features of another cluster are [0.5, 0.6, 0.4, 0.5, 10, 0.7], and the output score is 0.9, indicating a high possibility of anomaly.

[0087] When the anomaly score of a vehicle cluster exceeds the preset risk threshold (the threshold value range can be set from 0.7 to 0.9 and adjusted according to actual conditions), the system will generate a weighing anomaly alarm and send it to the process control module. For example, if the threshold is 0.8, a cluster with a score of 0.9 will trigger an alarm.

[0088] In a further embodiment, the process control module of the present invention includes three submodules:

[0089] The alarm recording submodule receives alarm information and records details, such as the list of vehicle IDs in the abnormal cluster (such as "Car 1, Car 2, Car 3"), the anomaly score (0.9), and the matching feature vector. These records are stored in the database for subsequent auditing.

[0090] The control submodule is linked to trigger control strategies based on alarm information, such as restricting the passage of abnormal vehicles (via electronic fence notification), strengthening monitoring (real-time GPS tracking), or notifying management personnel. Strategies can be dynamically adjusted based on the severity of the abnormality.

[0091] The manual review submodule pushes suspected abnormal clusters to the operations and maintenance platform, providing a visual interface displaying the trajectory, weight, and other information of the vehicles within the cluster for manual confirmation. After manual review, the status of the abnormal vehicles can be updated to ensure that false alarms are corrected.

[0092] like Figure 3 and Figure 4As shown, for a vehicle cluster that is determined to be abnormal, in some embodiments, suspected abnormal vehicles therein can be further identified. The specific steps are as follows: extract the feature vector of each vehicle from the abnormal cluster, for example [1, 20, 1697358922, 18.5, 0.3, 0.925, GPS trajectory]. Input the feature vector into the trained abnormal vehicle identification model and output the abnormality score (0 to 1) of each vehicle. The training process of the abnormal vehicle identification model of the present invention is similar to that of the cluster-level abnormality identification model: construct samples (feature vectors of each vehicle) based on historical data, manually mark whether it is abnormal, and use a supervised learning algorithm (such as random forest, the number of trees is recommended to be 100) for training.

[0093] A preset threshold is set (e.g., 0.6 to 0.8), and vehicles with scores above the threshold are marked as suspected anomalies. For example, if the threshold is 0.7 and the vehicle scores are 0.9, 0.8, and 0.5, respectively, the first two vehicles are suspected anomalies.

[0094] Through the above steps, the present invention can identify coordinated anomalies suspected of being a coordinated crime by multiple people within the overall transport behavior. Furthermore, through the abnormal vehicle identification model, it can further select abnormal vehicles within the cluster and, combined with various investigative methods such as manual questioning, further determine whether the abnormal cluster is organized behavior or individual vehicle violations.

[0095] Based on the above, the present invention further discloses a weighing abnormality alarm and full-process intelligent management and control system, including the following modules:

[0096] A vehicle feature construction module, configured to collect vehicle weighing behavior data and construct a vehicle feature vector based on the weighing behavior data;

[0097] The vehicle clustering module is used to cluster the vehicle feature vectors of all vehicles to obtain multiple vehicle clusters with similar behavior patterns;

[0098] A task matching feature extraction module is used to match the vehicle cluster with its corresponding transportation task and extract matching features based on the information of the vehicle cluster and the transportation task;

[0099] A cluster-level anomaly scoring module is used to input the matching features into a trained cluster-level anomaly recognition model and output an anomaly score for each vehicle cluster;

[0100] The abnormal alarm trigger module is used to output the weighing abnormal alarm information to the process control module when the abnormal score of a vehicle cluster exceeds the preset risk threshold.

[0101] To achieve truly implemented "end-to-end intelligent management and control", the present invention further builds a full-process engineering deployment system covering on-site perception, remote monitoring, data closure and abnormal feedback on the basis of the process control module, ensuring that the identified abnormal behaviors can be handled in a timely and efficient manner, and dynamic optimization and supervision can be achieved in long-term operation.

[0102] Take oil depot management as an example: First, the system supports an abnormal weight difference alarm mechanism, which intelligently monitors the weighing process before and after the vehicle is unloaded. When the vehicle is weighed for the first time, the system automatically calculates and records its gross weight and estimated tare weight. When the vehicle is weighed again after unloading, it automatically records the new tare weight and compares it with the previous value to determine whether it exceeds the preset loss error threshold (such as ±0.3 tons). Once the deviation is found to be out of limit, the system will immediately trigger an audible and visual alarm on site and remotely push it to the backend responsible person platform, prompting staff and drivers to conduct on-site verification and correction, thereby effectively preventing fraud and economic losses during the unloading process.

[0103] The system is designed to ensure that all pound value data cannot be tampered with. Weighing-related data (including gross weight, tare weight, timestamp, license plate, GPS trajectory, etc.) are automatically collected and encrypted and stored through the interface. It also supports docking with the enterprise oil depot management platform to achieve unified data management and cross-platform call analysis, providing a complete traceable link for subsequent audits, optimization and accountability.

[0104] To adapt to the needs of unmanned oil depots and nighttime operations, the system also supports 24-hour self-service oil unloading operations. It is equipped with smart cameras and voice interaction devices. Staff can view on-site video images in real time through the remote platform, and intervene and provide guidance immediately through voice terminals when abnormal operations or process deviations are found, effectively improving remote supervision efficiency.

[0105] During continuous operation, the system can conduct loss data analysis and modeling on the oil unloading behavior of each vehicle. By performing cluster statistics and trend analysis on the "gross weight-tare weight-net weight-loss" information in multiple weighing records, it can identify vehicles with abnormal long-term losses or abnormal trends at fixed unloading points, and assist management to intervene in investigations in advance and formulate rectification plans to achieve the goal of reducing and controlling losses.

[0106] To enhance visualization and audit assurance, the system automatically captures vehicle images every time a vehicle is weighed, and overlays and displays relevant data (such as time, gross weight, tare weight, net weight, loss, etc.), so that the image data and numerical data form a linked chain of evidence; at the same time, all weighing records and images are generated into uniquely numbered flow record entries in chronological order, which is convenient for on-demand query and export.

[0107] In addition, the system supports a remote planning and ordering mechanism. Oil depot managers can send weighing vehicle information to the oil depot terminal in advance through the intranet platform. Weigh operators can only review the issued plan and do not have any authority to add or modify vehicle information, further preventing the risks brought by the human addition or replacement of vehicles.

[0108] In special circumstances, such as when the vehicle identification camera fails, the system allows you to double-click the camera window to enlarge the video image, combine manual verification of the license plate and manual entry of the license plate number, to ensure the manual weighing operation process under emergency response and ensure the continuity and robustness of the system operation.

[0109] Finally, all data is integrated into a visualization platform, enabling tracking and auditing of the entire process, from order placement and vehicle weighing, anomaly identification, and alarm handling. Managers can readily access depot weighing records, alarm details, and unloading schedules, enabling a comprehensive, multi-dimensional, closed-loop management and control system, from single-vehicle identification to integrated transport group collaboration and ultimately business processes.

[0110] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A weighing abnormality alarm and full-process intelligent management and control method, characterized in that: The steps include: Collecting vehicle weighing behavior data and constructing a vehicle feature vector based on the weighing behavior data; Clustering the vehicle feature vectors of all vehicles to obtain multiple vehicle clusters with similar behavior patterns; Matching the vehicle cluster with its corresponding transport task, and extracting matching features based on information of the vehicle cluster and the transport task; Input the matching features into the trained cluster-level anomaly recognition model and output an anomaly score for each vehicle cluster; When the abnormal score of a vehicle cluster exceeds the preset risk threshold, the weighing abnormality alarm information is output to the process control module.

2. The method according to claim 1, characterized in that The weighing behavior data includes any one or more of the following: vehicle model, vehicle rated tonnage, weighing time, weighing weight, transportation route, loading ratio, and GPS track.

3. The method according to claim 1, characterized in that The matching features include any one or more of the following: the deviation ratio between the total loaded weight of the vehicle cluster and the nominal total weight of the transport task; The average trajectory similarity between the transport trajectory of each vehicle in the vehicle cluster and the preset trajectory of the transport task; The average overlap between the actual arrival time of the vehicles in the vehicle cluster and the specified execution time window of the transport task; The average proportion of the trajectory paths of the vehicles in the vehicle cluster covering the transportation task path; the standard deviation of the loaded weights of the vehicles in the vehicle cluster; The average loading rate of the vehicle cluster.

4. The method according to claim 1, characterized in that The cluster-level anomaly recognition model is a binary classification or regression scoring model trained based on supervised learning. The model uses multiple historically labeled normal transport clusters and abnormal transport clusters as training samples. The matching features are input and the output is a probability score of whether each vehicle cluster is abnormal transportation.

5. The method according to claim 4, characterized in that: The training process of the cluster-level anomaly recognition model includes the following steps: Construct multiple vehicle clusters based on historical transportation data. Each vehicle cluster is generated by a clustering algorithm and contains vehicle feature vectors of all vehicles in the cluster and their corresponding transportation task information. Manually label the vehicle clusters or perform system reconciliation labeling to obtain a label indicating whether each vehicle cluster is an abnormal transport cluster; Extract the matching features for each vehicle cluster as the input vector of the training sample, and use the label as the training output; The cluster-level anomaly recognition model is trained using a supervised learning method.

6. The method according to claim 1, characterized in that The clustering method is a density-based clustering algorithm or a hierarchical clustering algorithm based on a similarity threshold.

7. The method according to claim 1, characterized in that: The method further includes: after determining that a certain vehicle cluster is an abnormal cluster, identifying suspected abnormal vehicles from the abnormal cluster, the identification process including: Extracting the vehicle feature vector of each vehicle in the abnormal cluster; Inputting the vehicle feature vector into an abnormal vehicle identification model and outputting an abnormality score for each vehicle in the abnormal cluster; Based on the anomaly score, one or more vehicles in the vehicle cluster having an anomaly score higher than a preset threshold are determined as suspected abnormal vehicles.

8. The method according to claim 7, characterized in that: The training process of the abnormal vehicle identification model includes the following steps: A training sample set is constructed based on historical transportation records and transportation task reconciliation information. Each sample corresponds to the vehicle feature vector of a vehicle. Based on manual verification results, the vehicle is labeled to identify whether it is an abnormal vehicle; A supervised learning algorithm is used to train the abnormal vehicle recognition model.

9. The method according to claim 1, characterized in that: The process control module includes an alarm record submodule, a linkage management and control submodule, and a manual review submodule; The alarm recording submodule is used to receive and record the abnormal weighing alarm information and the IDs of all vehicles in the corresponding vehicle cluster; The linkage control submodule is used to trigger a control strategy for vehicles in an abnormal vehicle cluster after receiving abnormal alarm information; The manual review submodule is used to push suspected abnormal vehicle clusters to the operation and maintenance platform and provide a visual analysis interface for manual confirmation.

10. A weighing abnormality alarm and full-process intelligent management and control system, characterized in that: Includes the following modules: A vehicle feature construction module, configured to collect vehicle weighing behavior data and construct a vehicle feature vector based on the weighing behavior data; The vehicle clustering module is used to cluster the vehicle feature vectors of all vehicles to obtain multiple vehicle clusters with similar behavior patterns; A task matching feature extraction module is used to match the vehicle cluster with its corresponding transportation task and extract matching features based on the information of the vehicle cluster and the transportation task; A cluster-level anomaly scoring module is used to input the matching features into a trained cluster-level anomaly recognition model and output an anomaly score for each vehicle cluster; The abnormal alarm trigger module is used to output the weighing abnormal alarm information to the process control module when the abnormal score of a vehicle cluster exceeds the preset risk threshold.

Images