Truck Over-limit and Overload Classification and Grading Control System Based on Big Data Profiling Technology

Abstract

The present invention discloses a classification and grading control system for overloaded trucks based on big data profiling technology, belonging to the technical field of transportation management. Specifically, it includes: collecting and preprocessing the dynamic driving data of trucks from multiple data sources; deeply integrating the data by using a multi-modal feature fusion strategy, extracting multi-dimensional comprehensive feature information of trucks, and based on the multi-dimensional comprehensive feature information of trucks, constructing big data profiling templates for different types of trucks by using data mining and machine learning to accurately outline the characteristics of various trucks; based on the profiling, combining with the real-time dynamic driving data to judge in real time whether the truck is overloaded, accurately grading according to the overloading ratio, illegal frequency and harmful consequences, and implementing differential control for trucks with different grades, solving the problem of overloaded truck supervision and improving the transportation management efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of transportation management, and specifically relates to a classification and grading control system for overloaded trucks based on big data profiling technology. Background Art

[0002] Overloaded truck transportation not only causes serious damage to highway infrastructure, shortens the service life of highways, and increases maintenance costs, but also greatly affects traffic safety and is prone to traffic accidents. The traditional control method for overloaded trucks mainly relies on fixed or mobile weighing and detection stations, which has problems such as limited detection coverage, easy omission of inspections, and inability to dynamically and real-time grasp the transportation situation of trucks, making it difficult to achieve precise and efficient control.

[0003] For example, the Chinese patent application with the publication number CN116644304A discloses a method for extracting and classifying abnormal characteristics of truck bearings based on similarity, including: collecting basic data; the basic data includes the time-domain acceleration signal of the truck bearing; preprocessing the basic data and dividing the basic data into a training set and a test set; and converting the time-domain data in the training set and the test set into frequency-domain data; establishing a twin network feature extraction model; and in the established twin network feature extraction model, there is a feature calculation network and a twin network twin to it; the structures of the feature calculation network and the twin network are the same and the parameters are shared; using the twin network feature extraction model for feature extraction. This technical solution can intelligently and accurately complete the identification and analysis of abnormal truck bearings, helps to identify early faults of truck bearings, has a high accuracy and is real-time.

[0004] For example, the Chinese patent with the authorization announcement number CN112905578B discloses a method for identifying truck GPS trajectory stop points, including: based on the vehicle unique identifier ID and the truck GPS trajectory data with chaotic time series, cleaning the GPS trajectory data, that is, data screening, to eliminate duplicate and invalid data; based on the cleaned GPS trajectory data, classifying and clustering the GPS trajectory data with different vehicle unique identifier IDs in the same data file; based on the GPS trajectory data classified by vehicle ID, reordering the time series of the data; based on the processed GPS trajectory data, according to the actual research requirements for the stop time at the stop point, calculating the density value to determine the stop points in the truck GPS trajectory data; based on the determined stop point data, calculating the stop time of the vehicle at each stop point, and screening the stop points according to the stop time limit required by the actual research, and finally storing them as a local TXT text file.

[0005] The above existing technologies all have the following problems: narrow application scenarios, single data dimension; lack of systematic control measures. Summary of the Invention

[0006] In view of the deficiencies of the prior art, the present invention proposes a classification and grading control system for overloaded trucks based on big data profiling technology, which collects and preprocesses the dynamic driving data of trucks from multiple data sources; uses a multi-modal feature fusion strategy to deeply integrate the data, extracts the multi-dimensional comprehensive feature information of trucks, and based on the multi-dimensional comprehensive feature information of trucks, uses data mining and machine learning to construct big data profiling templates for different types of trucks, accurately outlining the characteristics of various trucks; based on the profiling, combines the real-time driving data to judge in real time whether the truck is overloaded, and accurately grades according to the overloading ratio, illegal frequency and harmful consequences, and implements differential control for trucks with different grades, solving the problem of overloaded truck supervision and improving the efficiency of transportation management.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A classification and grading control system for overloaded trucks based on big data profiling technology includes: a data analysis module, a profiling construction module, a judgment and grading module, and a control and execution module;

[0009] The data analysis module includes a data collection unit and a fusion unit. The data collection unit is used to collect the dynamic driving data of trucks through sensors; the fusion unit is configured with a multi-modal feature fusion strategy, and the multi-modal feature fusion strategy is used to deeply fuse the preprocessed dynamic driving data of trucks to extract the multi-dimensional comprehensive feature information of trucks;

[0010] The profiling construction module includes a clustering analysis unit and a profiling template generation unit. The clustering analysis unit is configured with an improved clustering strategy, and the improved clustering strategy is used to judge whether the current point in the fused truck data is a core point for clustering; the profiling template generation unit is used to generate big data profiling templates for each truck group;

[0011] The judgment and grading module is used to comprehensively judge whether the truck has overloaded behavior by combining the real-time collected dynamic driving data of the truck and the big data profiling, and grade according to the preset grading rules;

[0012] The control and execution module includes a measure formulation unit, and the measure formulation unit is used to formulate differential control measures according to the grading results;

[0013] The fusion unit adopts a multi-modal feature fusion strategy, and the specific steps of the multi-modal feature fusion strategy include:

[0014] A1: Obtain the preprocessed dynamic driving data of trucks from the distributed data storage nodes, and accurately divide the preprocessed dynamic driving data of trucks into five modal groups according to the data source and deep feature attributes;

[0015] A2: Using a distributed computing framework, divide the data of each modality group into N data blocks, and assign the task of normalizing each modality group data to different computing nodes;

[0016] A3: On each computing node assigned with tasks, use a normalization method to calculate the respective responsible data blocks, and summarize the results back to the master node of the distributed computing framework through the communication and synchronization mechanism of the distributed computing framework. The formula is:

[0017] ;

[0018] Wherein, represents the truck dynamic driving data in the j-th data block of the i-th modality group after normalization, represents the truck dynamic driving data in the j-th data block of the i-th modality group, represents the corresponding minimum value in, represents the corresponding maximum value in, a and b represent scaling factors, c represents the coefficient of non-linear transformation, represents a positive number, represents a non-linear function;

[0019] A4: Organize the normalized data of each modality group according to the feature dimensions to form the normalized feature vectors of each modality group , wherein, represents the truck dynamic driving data in the N-th data block of the i-th modality group after normalization;

[0020] The specific steps of the multi-modal feature fusion strategy further include:

[0021] A5: According to the normalized feature vectors of each modality group, calculate the correlation between pairwise modality groups and generate a multi-dimensional correlation matrix , the formula is:

[0022] ;

[0023] Wherein, represents the n-th truck dynamic driving data in the i-th modality group after normalization, n represents the number of truck dynamic driving data, and , represents the number of truck dynamic driving data in each data block, represents the correlation between the e-th modality group and the r-th modality group, represents the weighting factor of the k-th truck dynamic driving data in the normalized e-th modality group and r-th modality group, and respectively represent the k-th truck dynamic driving data in the e-th and r-th modal groups after normalization, and respectively represent the mean values of the truck dynamic driving data in the e-th and r-th modal groups after normalization, and respectively represent the offsets of the k-th truck dynamic driving data in the e-th and r-th modal groups after normalization, where r and e represent two different modal groups, represents the correlation between the i-th modal group and the (i - 1)-th group;

[0024] The specific steps of the multi-modal feature fusion strategy further include:

[0025] A6: Based on the multi-dimensional correlation matrix , introduce the attention mechanism, and combine the output of the attention mechanism with the business priority rules, and calculate the real-time weights of different modalities through the improved Softmax function to form a dynamic weight vector , where, represents the importance score of the adjusted i-th modal group, represents the regularization parameter, u represents the bias term, and v represents the weight adjustment parameter;

[0026] A7: Weighted sum the feature vectors of each modality according to the dynamic weight vector to obtain the final fused truck data, and store it in the feature extraction input buffer.

[0027] Specifically, the specific steps of A6 include:

[0028] A6.1: Obtain the multi-dimensional correlation matrix , and introduce the attention mechanism;

[0029] A6.2: Use the elements in as the input of the attention mechanism, and through , obtain the attention score vector , where, represents the importance score of the i-th modal group, represents the attention mechanism function;

[0030] A6.3: According to the business requirements, define the priority rules of different modalities, and obtain the business priority score vector according to the priority rules of different modalities. The priority rules are based on the scoring results of the reliability, importance, and real-time factors of the modality for priority sorting, where, represents the priority score vector of the i-th modal group.

[0031] Specifically, the specific steps of A6 also include:

[0032] A6.4: According to the formula , and get the adjusted score vector ,in, represents the importance score of the i-th modal group after adjustment;

[0033] A6.5: The adjusted score vector Input into the improved Softmax function, calculate the real-time weights of different modes, and generate a dynamic weight vector ,in, represents the improved Softmax function, represents the regularization parameter.

[0034] Specifically, the cluster analysis unit adopts an improved clustering strategy, and the specific steps of the improved clustering strategy include:

[0035] B1: Extract the final fused truck data obtained from A7 from the buffer , according to the clustering requirements, set the judgment threshold h of the core point, where, represents the qth final fused truck data, q represents the number of final fused truck data;

[0036] B2: For Each data point in , calculate the density around each data point ;

[0037] B3: The density around each data point Compare with h;

[0038] like , then determine the data point As the core point;

[0039] B4: Taking the core point as the center, the surrounding data points are classified into the same cluster according to the set h;

[0040] B5: For the formed clusters, continue to look for clusters that are adjacent to the formed clusters and satisfy of data points and will satisfy The data points are classified into this cluster until no more data points that meet the conditions are found;

[0041] B6: Extract characteristic information of each truck group from the clustering results, and fill the extracted characteristic information into a preset big data portrait template to generate a big data portrait template for each truck group. The characteristic information of the truck group includes driving route, speed distribution, and stop time.

[0042] Specifically, the five modal groups in A1 include: basic vehicle attribute data, transportation behavior characteristic data, violation record data, weather condition data and cargo information; the basic vehicle attribute data include vehicle model, age, load tonnage, and company to which it belongs; the transportation behavior characteristic data include transportation route, transportation frequency, commonly loaded cargo type, and average driving speed; the violation record data include the number of over-limit and overload times, location, and severity; the weather condition data include rainy days and strong winds; and the cargo information includes special cargo attributes.

[0043] Specifically, the judgment and classification module includes a behavior judgment unit, a classification rule setting unit, and a classification processing unit;

[0044] The behavior judgment unit is used to judge the overload behavior of the truck by using the big data portrait template and the real-time dynamic driving data of the truck;

[0045] The classification rule setting unit is used to set the classification rules according to the overload and overlimit ratio, the frequency of historical violations, and the harmful consequences caused;

[0046] The grading processing unit is used to perform grading processing on overloaded trucks according to grading rules.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] 1. The present invention proposes a classification and grading control system for overloaded trucks based on big data profiling technology, and optimizes and improves the architecture, operating steps and processes. The system has the advantages of simple processes, low investment and operating costs, and low production work costs.

[0049] 2. The present invention proposes a classification and grading control system for overloaded trucks based on big data profiling technology. By collecting dynamic driving data of trucks from multiple sources, it not only ensures that the data is real-time and accurate, but also provides a basis for subsequent analysis after pre-processing; it uses a multimodal feature fusion strategy to deeply mine data, integrates multi-dimensional comprehensive features such as basic vehicle attributes, transportation behavior, violations of laws and regulations, weather conditions, cargo information, etc., to outline the overall picture of the truck in all directions, and effectively improve the ability to accurately understand the operating status of the truck.

[0050] 3. The present invention proposes a classification and grading control system for over-limit and over-load trucks based on big data profiling technology. By means of data mining and machine learning, a big data profiling template is constructed, and combined with real-time driving data, over-limit and over-load judgment and grading are carried out, making the control more targeted. It can be finely graded according to the over-limit and over-load degree of trucks, historical violation situations, harmful consequences, etc., accurately divide the categories of mild, moderate, and severe over-limit and over-load, and then formulate differential control measures for different grades, forming a complete closed loop from source prevention, process supervision to post-punishment, effectively guaranteeing highway transportation safety, standardizing the logistics market order, and reducing safety risks while improving traffic management efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is the architecture diagram of the classification and grading control system for over-limit and over-load trucks based on big data profiling technology of the present invention;

[0052] Figure 2 It is the principle flow chart of the classification and grading control system for over-limit and over-load trucks based on big data profiling technology of the present invention;

[0053] Figure 3 It is the flow chart of the algorithm implementation of the fusion unit of the classification and grading control system for over-limit and over-load trucks based on big data profiling technology of the present invention;

[0054] Figure 4 It is the flow chart of the algorithm implementation of the clustering analysis unit of the classification and grading control system for over-limit and over-load trucks based on big data profiling technology of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0055] Embodiment 1

[0056] Please refer to Figure 1 and Figure 2 , an embodiment provided by the present invention: A classification and grading control system for over-limit and over-load trucks based on big data profiling technology, including:

[0057] A data analysis module, a profiling construction module, a judgment and grading module, and a control and execution module;

[0058] The data analysis module is used to collect the dynamic driving data of trucks from multiple data sources and preprocess the dynamic driving data of trucks to ensure the accuracy, integrity, and consistency of the data;

[0059] The profiling construction module is used to use data mining algorithms and machine learning models to construct big data profiling templates for different types of trucks according to the extracted multi-dimensional comprehensive feature information of trucks;

[0060] The judgment and grading module is used to combine the real-time collected dynamic driving data of trucks and the big data profile, comprehensively judge whether the truck has over-limit and over-load behaviors, and grade according to the preset grading rules;

[0061] A control and execution module, which is used to formulate and adopt differential control measures for overloaded trucks with different classifications and levels.

[0062] The data analysis module includes: a data collection unit, a preprocessing unit, and a fusion unit;

[0063] The data collection unit is used to collect the dynamic driving data of trucks through sensors;

[0064] The preprocessing unit is used to preprocess the collected dynamic driving data of trucks, including removing duplicates, filling missing values, and processing outliers in the collected dynamic driving data of trucks, etc., to improve the data quality;

[0065] The fusion unit is used to deeply fuse the preprocessed dynamic driving data of trucks by adopting a multi-modal feature fusion strategy according to the preprocessed dynamic driving data of trucks, and extract multi-dimensional comprehensive feature information of trucks.

[0066] The portrait construction module includes: a clustering analysis unit and a portrait template generation unit;

[0067] The clustering analysis unit is used to perform clustering analysis on the multi-dimensional comprehensive feature information of trucks to identify different types of truck groups;

[0068] The portrait template generation unit is used to generate a representative big data portrait template for each truck group, including a feature vector and a typical feature description.

[0069] The judgment and grading module includes: a behavior judgment unit, a grading rule setting unit, and a grading processing unit;

[0070] The behavior judgment unit is used to judge the overloading behavior of trucks by using the big data portrait template and the real-time dynamic driving data of trucks;

[0071] The grading rule setting unit is used to set grading rules according to factors such as the overloading ratio, the frequency of historical violations, and the resulting harmful consequences;

[0072] The grading processing unit is used to perform grading processing on overloaded trucks according to the grading rules, providing a basis for subsequent control measures.

[0073] The control and execution module includes: a measure formulation unit, a measure execution unit, and an evaluation and feedback unit;

[0074] The measure formulation unit is used to formulate differential control measures according to the grading results, such as warnings, fines, and revocation of operation permits;

[0075] The measure execution unit is used to implement the formulated differential control measures, including notifying truck drivers and imposing penalties;

[0076] The evaluation and feedback unit is used to evaluate the effectiveness of the implementation of control measures, and to adjust and optimize the measures based on the evaluation results.

[0077] In summary, the overall implementation process of the truck overload classification and grading control system based on big data profiling technology includes:

[0078] Step S1: Collect dynamic driving data of trucks from multiple data sources, including weighing data from highway toll stations, detection data from overload control stations, Beidou satellite navigation system or global positioning system trajectory data of trucks, registration information from vehicle management offices, transportation order information from logistics companies, and dynamic data such as axle weight, gross weight, vehicle speed, and tire pressure during the driving of trucks, to ensure the real-time and accuracy of the data, and pre-process the collected dynamic driving data of trucks;

[0079] Step S2: A multimodal feature fusion strategy is used to deeply fuse the pre-processed dynamic driving data of the truck to extract multi-dimensional comprehensive feature information of the truck, which includes basic vehicle attributes, such as vehicle model, age, load tonnage, and company; transportation behavior characteristics, such as transportation route, transportation frequency, common cargo type, and average driving speed; violation records, including the number, location, and severity of overloads; real-time weather conditions, such as the need to reduce speed on slippery roads in rainy days and the impact of strong winds on the stability of high-box trucks; cargo information, such as special cargo attributes such as fragile goods, flammable and explosive goods;

[0080] Step S3: using data mining algorithms and machine learning models, taking the multi-dimensional comprehensive feature information of trucks extracted in step S2 as input, performing cluster analysis according to similar features, and constructing big data portrait templates of different types of trucks, wherein each portrait template contains a set of representative feature vectors to describe the typical features of this type of truck, and the data mining algorithms and machine learning models are prior art contents in this field, and are not the inventive solutions of the present application, and are not elaborated here;

[0081] Step S4: Based on the truck big data portrait and the real-time collected dynamic driving data of the truck, a comprehensive judgment is made as to whether the truck has overloaded or not. For the trucks that are determined to be overloaded or not, the trucks are classified according to the overload ratio, the frequency of historical violations, the harmful consequences caused, and other factors through the preset classification rules. For example, a truck that is overloaded or not by 30% and violates the law for the first time is classified as a light overloaded truck, a truck that is overloaded or not by 30% and violates the law 2-3 times within a year is classified as a moderate overloaded truck, and a truck that is overloaded or not by 50% or more or causes serious traffic congestion or road damage many times is classified as a severe overloaded truck.

[0082] Furthermore, the specific steps of step S4 include:

[0083] (1) Data collection and integration:

[0084] Collect big data portrait information of trucks, including historical driving data, load records, and vehicle model information; at the same time, collect real-time driving data of trucks, such as current load, speed, location, etc.;

[0085] (2) Judgment of over-limit and overload:

[0086] Compare the actual load of the truck with the approved load, calculate the overload rate by (actual load - approved load) / approved load × 100%, and determine whether there is overloading behavior based on the overload rate;

[0087] (3) Setting of classification rules:

[0088] Set classification rules, which can be based on factors such as the proportion of overload and overlimit, the frequency of historical violations, and the harmful consequences caused;

[0089] (4) Comprehensive evaluation and classification:

[0090] Based on the real-time collected dynamic driving data and big data portrait information of trucks, the overload and overloading behaviors of trucks are comprehensively evaluated, and trucks are classified into different levels according to the set classification rules;

[0091] (5) Result output and application:

[0092] The classification results are output, and corresponding measures are taken according to different levels, such as warnings, fines, and temporary vehicle seizure. The classification results are used for subsequent supervision and management to improve the safety and compliance of truck transportation.

[0093] Step S5: Formulate and adopt differentiated control measures for overloaded trucks of different classifications and grades output in step S4;

[0094] For trucks that are slightly overloaded, early warning information will be sent to the driver through the vehicle terminal, reminding him to go to the nearest designated unloading point to unload the goods. The illegal information will be recorded and the company will be notified to strengthen safety education.

[0095] For moderately overloaded trucks, in addition to the above-mentioned warnings and notifications, their driving trajectory will be tracked in real time, and they will be directed to nearby overload control stations for processing, and fines and points will be imposed in accordance with the law, and the companies will be required to make rectifications within a time limit;

[0096] For severely overloaded trucks, a joint law enforcement mechanism will be immediately activated to intercept them, and the cargo will be forced to be unloaded and the vehicles will be seized. Penalties will be imposed on the drivers and their companies, including high fines, suspension of business, revocation of relevant qualifications, etc. The companies will also be included in the key supervision list, and the frequency of daily inspections will be increased.

[0097] Example 2

[0098] See also Figure 3 In this embodiment, the fusion unit adopts a multimodal feature fusion strategy. The specific steps of the multimodal feature fusion strategy include:

[0099] A1: Obtain pre-processed truck dynamic driving data from the distributed data storage node, and accurately divide the pre-processed truck dynamic driving data into five modal groups based on the data source and deep feature attributes;

[0100] Among them, data sources, such as data from different sensor types and different acquisition devices; deep feature attributes, such as speed-related features and load-related features.

[0101] Furthermore, the specific steps of accurately dividing the pre-processed truck dynamic driving data into five modal groups include:

[0102] (1) Data feature combing: Comprehensively comb the pre-processed truck dynamic driving data to identify all potential data features;

[0103] (2) Formulate classification rules for modal groups: Based on the five known modal groups and the specific data categories they contain, formulate precise classification rules:

[0104] For the vehicle basic attribute data modal group: set rules, as long as the data features directly describe the inherent characteristics of the vehicle itself, such as clearly marked as vehicle model, age, load tonnage, company to which it belongs, etc., it will be classified into this modal group. For example, in a data record: Dongfeng Tianlong heavy truck, age 5 years, load tonnage 30 tons, company A Logistics, all elements in it should be classified into this modal group;

[0105] For the transport behavior characteristic data modal group: if the data reflects the behavior patterns and dynamic conditions of trucks during operation, such as the transport route is a series of geographical coordinates or road section names, the transport frequency is the number of trips within a certain period of time, the common cargo type is the cargo name or classification, and the average driving speed is the speed value, they are all classified into this modal group. For example, if the transport frequency this month is 15 times, the common cargo is steel, the transport route is from City B to City C, and the average driving speed is 60km / h, these data should be placed in this modal group;

[0106] Regarding the violation record data modal group: when the data involves past violations of laws and regulations by trucks, and clearly points to the number of violations related to overloading, the location of occurrence, and the severity, where the severity is expressed in terms of the amount of fines, deduction points, or the level of assessment by law enforcement agencies, it is classified into the violation record data modal group; for example, on the D highway section, there were 3 violations of overloading, and the severity was medium, and this type of information belongs to this modal group;

[0107] For the weather condition data modality group: The data is sourced from a meteorological monitoring system and describes the weather phenomena encountered during the truck's journey, such as weather labels like rainy days and strong winds, along with the corresponding time and location ranges. Such data is classified into this modality group. For example, on July 10, 2023, in the section from City B to City C, when it rained, this piece of data belongs to this modality group;

[0108] For the cargo information modality group: As long as the data pertains to the special properties, categories, etc. of the cargo itself, such as descriptions of special cargo attributes like fragile goods, inflammable and explosive goods, or annotations related to special cargo in a detailed cargo list, it should be classified into the cargo information modality group. For example, if the goods transported this time include fragile porcelain, this information enters this modality group;

[0109] (3)Data classification and preliminary grouping:

[0110] According to the established classification rules, each piece of preprocessed truck dynamic driving data is traversed. Through programming means, such as using the Pandas library in Python for data processing and applying conditional judgment statements, for each data feature, determine the modality group to which it should belong. For example, for a comprehensive data record containing multiple features, each feature is checked in turn. If it conforms to the classification rules of vehicle basic attribute data, add this feature to the temporary storage structure of vehicle basic attribute data; if it meets the rules of transportation behavior feature data, store it in the corresponding storage location of transportation behavior feature data, and so on, to complete the preliminary grouping work;

[0111] (4)Group verification and adjustment:

[0112] After the preliminary grouping is completed, each modality group is verified. On the one hand, check whether the data within each modality group is consistent and logical to avoid incorrect classification. For example, in the vehicle basic attribute data modality group, if a feature like the transportation speed of 80 km / h this time is found, it obviously does not conform to the definition of this modality group and should be reclassified into the transportation behavior feature data modality group; on the other hand, ensure that the five modality groups can comprehensively cover all preprocessed truck dynamic driving data without omission or duplicate classification. If problems are found, adjust the classification strategy in a timely manner and reassign the incorrectly classified data until the grouping result is accurate.

[0113] A2: Using a distributed computing framework, divide the data of each modality group into N data blocks and assign the task of normalizing the data of each modality group to different computing nodes;

[0114] Furthermore, the specific steps of A2 include:

[0115] (1) In a distributed computing framework, tasks of normalization processing are assigned to different computing nodes. Each computing node is responsible for normalizing one or more data blocks, and it is ensured that each computing node has the resources and environment required for performing normalization processing.

[0116] (2) The divided data blocks are distributed to the corresponding computing nodes through a distributed file system. Herein, the distributed file system is the prior art content in this field and is not the creative solution of this application, so it will not be elaborated herein.

[0117] A3: On each computing node assigned with tasks, the normalization method is used to calculate the respective responsible data blocks, and the results are summarized back to the master node of the distributed computing framework through the communication and synchronization mechanism of the distributed computing framework. The formula is:

[0118] ;

[0119] Wherein, represents the truck dynamic driving data in the j-th data block of the i-th modal group after normalization, represents the truck dynamic driving data in the j-th data block of the i-th modal group, represents the corresponding minimum value in, represents the corresponding maximum value in, a and b represent scaling factors, c represents the coefficient of non-linear transformation, represents a positive number, represents a non-linear function;

[0120] A4: The normalized data of each modal group are sorted according to the feature dimension to form the normalized feature vector of each modal group , wherein, represents the truck dynamic driving data in the N-th data block of the i-th modal group after normalization;

[0121] A5: According to the normalized feature vectors of each modal group, calculate the correlation between pairwise modal groups and generate a multi-dimensional correlation matrix , the formula is:

[0122] ;

[0123] Wherein, represents the n-th truck dynamic driving data in the i-th modal group after normalization, n represents the number of truck dynamic driving data, and , represents the number of truck dynamic driving data in each data block, Indicates the correlation between the e-th modal group and the r-th modal group. Indicates the weighting factor of the k-th truck dynamic driving data in the normalized e-th modal group and r-th modal group. And respectively indicate the k-th truck dynamic driving data in the normalized e-th modal group and r-th modal group. And respectively indicate the mean values of the truck dynamic driving data in the normalized e-th modal group and r-th modal group. And respectively indicate the offsets of the k-th truck dynamic driving data in the normalized e-th modal group and r-th modal group. r and e represent two different modal groups. Indicates the correlation between the i-th modal group and the (i - 1)-th group.

[0124] A6: Based on the multi-dimensional correlation matrix , introduce the attention mechanism, and combine the output of the attention mechanism with the business priority rules, and calculate the real-time weights of different modalities through the improved Softmax function to form a dynamic weight vector , where indicates the importance score of the adjusted i-th modal group. indicates the regularization parameter, u represents the bias term, and v represents the weight adjustment parameter.

[0125] A7: Weighted sum the feature vectors of each modality according to the dynamic weight vector to obtain the final fused truck data, and store it in the feature extraction input buffer.

[0126] In the present invention, it can be seen from step A6 that the dynamic weight vector is , perform a weighted sum on the dynamic weight vector: , for perform discretization to obtain the final fused truck data , indicates the weighting coefficient of the dynamic weight vector. indicates the weighted sum of the dynamic weight vector.

[0127] The specific steps of A6 include:

[0128] A6.1: Obtain the multi-dimensional correlation matrix , and introduce the attention mechanism. Among them, the attention mechanism is the prior art content in the field and is not the creative solution of this application, so it will not be elaborated here.

[0129] A6.2: Take the elements in as the input of the attention mechanism, and through , obtain the attention score vector , where represents the importance score of the i-th modality group, represents the attention mechanism function;

[0130] A6.3: According to business requirements, define the priority rules for different modalities. According to the priority rules of different modalities, obtain the business priority score vector , and the priority rules are sorted based on the scoring results of factors such as the reliability, importance, and real-time nature of the modality. Among them, represents the priority score vector of the i-th modality group;

[0131] A6.4: According to the formula , obtain the adjusted score vector , where represents the adjusted importance score of the i-th modality group;

[0132] A6.5: Input the adjusted score vector into the improved Softmax function to calculate the real-time weights of different modalities and generate the dynamic weight vector , and the calculation formula of the dynamic weight vector is:

[0133] ;

[0134] Among them, represents the improved Softmax function, represents the regularization parameter, represents the regularization term of the i-th modality group, represents the bias term of the i-th modality group, represents the weight adjustment parameter of the i-th modality group, represents the exponential function.

[0135] Embodiment 3

[0136] Please refer to Figure 4 . In this embodiment, the clustering analysis unit adopts an improved clustering strategy. The specific steps of the improved clustering strategy include:

[0137] B1: Extract the final fused truck data obtained in A7 from the buffer , and according to the clustering requirements, set the judgment threshold h of the core point. Among them, represents the q-th final fused truck data, and q represents the number of the final fused truck data;

[0138] B2: For each data point in , calculate the density around each data point ;

[0139] In the present invention, the specific steps of calculating the surrounding density include:

[0140] (1) Setting parameters: First, you need to set two key parameters of the DBSCAN algorithm: neighborhood radius and minimum number of points. , wherein the DBSCAN algorithm is a prior art in the art and is not an inventive solution of the present application, and will not be described in detail here;

[0141] (2) Calculate the neighborhood: For each data point in the data set, calculate the number of data points in its neighborhood. This is usually achieved by traversing the data set and checking whether the distance between each data point and every other data point is less than or equal to the neighborhood radius.

[0142] (3) Determine the core point: If the neighborhood of a data point contains at least data points, then the data point is considered a core point, which means that the density around the data point is high enough to form the core of a cluster.

[0143] B3: The surrounding density of each data point Compare with h;

[0144] like , then determine the data point As the core point;

[0145] B4: Taking the core point as the center, the surrounding data points are classified into the same cluster according to the set h;

[0146] B5: For the formed clusters, continue to look for clusters that are adjacent to the formed clusters and satisfy of data points and will satisfy The data points are classified into this cluster until no more data points that meet the conditions are found;

[0147] B6: Extract characteristic information of each truck group from the clustering results, and fill the extracted characteristic information into a preset big data portrait template to generate a big data portrait template for each truck group. The characteristic information of the truck group includes driving route, speed distribution, and stop time.

[0148] Furthermore, the specific steps of B6 include:

[0149] (1) Analyze the clustering results and identify different truck groups, i.e., clusters;

[0150] (2) For each truck group, extract its characteristic information, which may include:

[0151] Driving route: the main driving route or area of the truck group;

[0152] Travel time: the travel time distribution of truck groups, such as peak hours, off-peak hours, etc.;

[0153] Dwell time: the dwell time of truck groups at different locations;

[0154] Cargo type: the type of cargo transported by the truck group;

[0155] Speed distribution: the speed distribution of truck groups;

[0156] (3) Design a preset big data portrait template, which should contain fields or areas that can display the characteristic information of the truck group;

[0157] (4) Fill the extracted feature information into the corresponding fields or areas according to the format of the preset big data portrait template to ensure the accuracy and completeness of the information so as to generate a meaningful big data portrait;

[0158] (5) Generate a big data portrait template for each truck group based on the entered feature information;

[0159] (6) Verify the generated big data portrait template to ensure that it can accurately reflect the characteristic information of the truck group.

[0160] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in the field may also change, modify, replace and modify the above-mentioned embodiments without departing from the purpose and scope of protection of the present invention, and all of these are within the protection of the present invention.

[0161] If the disclosed technical solution involves personal information, the product using the disclosed technical solution has clearly informed the personal information processing rules and obtained the individual's voluntary consent before processing the personal information. If the disclosed technical solution involves sensitive personal information, the product using the disclosed technical solution has obtained the individual's separate consent before processing the sensitive personal information, and at the same time meets the "explicit consent" requirement. For example, on personal information collection devices such as cameras, clear and prominent signs are set to inform that the personal information collection scope has been entered and personal information will be collected. If the individual voluntarily enters the collection scope, it is deemed that he or she agrees to the collection of his or her personal information; or on the device that processes personal information, when the personal information processing rules are notified by obvious signs / information, the individual's authorization is obtained through pop-up information or by asking the individual to upload his or her personal information; among them, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the type of personal information processed.

Claims

1. A truck over-limit and over-load classification and grading control system based on big data profiling technology, characterized in that, include: Data analysis module, portrait construction module, judgment and classification module, control and execution module; The data analysis module includes a data acquisition unit and a fusion unit. The data acquisition unit is used to collect dynamic driving data of trucks through sensors. The fusion unit is configured with a multimodal feature fusion strategy. The multimodal feature fusion strategy is used to deeply fuse the pre-processed dynamic driving data of trucks to extract multi-dimensional comprehensive feature information of trucks. The portrait construction module includes a cluster analysis unit and a portrait template generation unit. The cluster analysis unit is configured with an improved clustering strategy, and the improved clustering strategy is used to determine whether the current point in the fused truck data is a core point and perform clustering; The portrait template generating unit is used to generate a big data portrait template for each truck group; The judgment and classification module is used to combine the real-time collected dynamic driving data of the truck and the big data portrait to comprehensively judge whether the truck has overloaded or overloaded behavior, and classify it according to the preset classification rules; The control and execution module includes a measure formulation unit, which is used to formulate differentiated control measures according to the classification results; The fusion unit adopts a multimodal feature fusion strategy, and the specific steps of the multimodal feature fusion strategy include: A1: Obtain pre-processed truck dynamic driving data from the distributed data storage node, and accurately divide the pre-processed truck dynamic driving data into five modal groups based on the data source and deep feature attributes; A2: Using the distributed computing framework, the data of each modality group is divided into N data blocks, and the task of normalizing the data of each modality group is assigned to different computing nodes; A3: On each computing node assigned to the task, the normalization method is used to calculate the data block for each node, and the results are summarized back to the master node of the distributed computing framework through the communication and synchronization mechanism of the distributed computing framework. The formula is: ； Among them, represents the dynamic driving data of trucks in the j-th data block of the i-th modal group after normalization, represents the dynamic driving data of trucks in the j-th data block of the i-th modal group, represents the corresponding minimum value in represents the corresponding maximum value in, a and b represent scaling factors, and c represents the coefficient of the non-linear transformation, represents a positive number, represents a non-linear function; A4: Organize the normalized data of each modality group according to the feature dimensions to form the normalized feature vectors of each modality group , where represents the truck dynamic driving data in the Nth data block of the ith modality group after normalization; A5: According to the normalized eigenvectors of each modality group , calculate the correlation between pairwise modality groups and generate a multi-dimensional correlation matrix , and the formula is: ； Among them, represents the nth truck dynamic driving data in the i-th normalized modal group, where n represents the number of truck dynamic driving data, and , represents the number of truck dynamic driving data in each data block, represents the correlation between the e-th modal group and the r-th modal group, represents the weighting factor of the k-th truck dynamic driving data in the e-th and r-th normalized modal groups, and respectively represent the k-th truck dynamic driving data in the e-th and r-th normalized modal groups, and respectively represent the means of the truck dynamic driving data in the e-th and r-th normalized modal groups, and respectively represent the offsets of the k-th truck dynamic driving data in the e-th and r-th normalized modal groups. r and e represent two different modal groups, represents the correlation between the i-th modal group and the (i - 1)-th group; The specific steps of the multimodal feature fusion strategy also include: A6: Based on the multi-dimensional correlation matrix , introduce the attention mechanism, combine the output of the attention mechanism with the business priority rules, and calculate the real-time weights of different modalities through the improved Softmax function to form a dynamic weight vector , where represents the importance score of the adjusted i-th modality group, represents the regularization parameter, u represents the bias term, and v represents the weight adjustment parameter; A7: Perform weighted summation on the feature vectors of each mode according to the dynamic weight vector to obtain the final fused truck data, and store it in the feature extraction input buffer; The cluster analysis unit adopts an improved clustering strategy, and the specific steps of the improved clustering strategy include: B1: Extract the final fused truck data obtained from A7 from the buffer , according to the clustering requirements, set the judgment threshold h for the core points, where represents the q-th final fused truck data, and q represents the number of the final fused truck data; B2: For each data point in , calculate the density around each data point ; B3: Compare the density around each data point with h; If , then determine that this data point is a core point; B4: Taking the core point as the center, the surrounding data points are classified into the same cluster according to the set h; B5: For the formed clusters, continue to search for data points that are adjacent to the formed clusters and satisfy , and classify the data points that satisfy into this cluster until no more data points that meet the conditions can be found; B6: Extracting characteristic information of each truck group from the clustering results, and filling the extracted characteristic information into a preset big data portrait template to generate a big data portrait template for each truck group, wherein the characteristic information of the truck group includes driving route, speed distribution, and dwell time; The judgment and classification module is used to comprehensively judge whether there is overload or overlimit behavior based on the big data portrait information of the truck and the real-time collection of the truck's driving data, and classify the truck into different levels according to the set classification rules; The real-time collection of truck driving data includes current load; The preset classification rules include the proportion of over-limit and overload, the frequency of historical violations of laws and regulations, and the harmful consequences caused.

2. The truck over-limit and over-load classification and grading control system based on big data profiling technology according to claim 1, wherein, The specific steps of A6 include: A6.1: Obtain a multi-dimensional correlation matrix , and introduce an attention mechanism; A6.2: Take the elements in as the input of the attention mechanism, and through , obtain the attention score vector , where represents the importance score of the i-th modality group, and represents the attention mechanism function; A6.3: Define the priority rules for different modalities according to business requirements, and obtain the business priority score vector according to the priority rules of different modalities. , where the priority rules are sorted according to the scoring results of the reliability, importance, and real-time factors of the modality. Among them, represents the priority score vector of the i-th modality group.

3. The truck over-limit and over-load classification and grading control system based on big data profiling technology according to claim 2, characterized in that The specific steps of the above-mentioned A6 further include: A6.4: According to the formula , the adjusted score vector is obtained, where represents the importance score of the i-th adjusted modality group; A6.5: Input the adjusted score vector into the improved Softmax function to calculate the real-time weights of different modalities and generate a dynamic weight vector , where represents the improved Softmax function, represents the regularization parameter.

4. The truck over-limit and over-load classification and grading control system based on big data profiling technology according to claim 3, wherein, The five modal groups in the above-mentioned A1 include: vehicle basic attribute data, transportation behavior characteristic data, violation record data, weather condition data, and cargo information; the vehicle basic attribute data includes vehicle type, vehicle age, load capacity tonnage, and affiliated enterprise; the transportation behavior characteristic data includes transportation route, transportation frequency, frequently carried cargo type, and average driving speed; the violation record data includes the number of over-limit and over-load, location, and severity; the weather condition data includes rainy days and strong winds; the cargo information includes special cargo attributes.

Images