A method and system for constructing key vehicle travel profile and risk assessment

Abstract

The present application relates to the technical field of intelligent transportation systems, and in particular to a method and system for constructing a travel portrait and risk assessment of key vehicles, which includes the following steps: collecting multi-source data and preprocessing; integrating a number of multi-source data into a unified database; fusing a number of multi-source data in the unified database to obtain a number of data subjects, and aligning the data subjects to obtain association relationships; extracting key feature information and integrating them to construct a feature candidate library; constructing a label system, and determining the type of portrait label corresponding to key vehicles based on the label system; setting multiple risk assessment dimensions, and constructing a risk assessment score result containing multiple risk assessment dimensions based on a number of feature candidate libraries combined with a risk assessment algorithm; and constructing a travel portrait of key vehicles in combination with the portrait label type and the risk assessment score result. The present application has the effect of improving the safety management capabilities of key vehicles.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent transportation systems, and in particular to a method and system for constructing a key vehicle travel profile and risk assessment. Background Art

[0002] Some types of transport and operating vehicles often have greater driving risks than ordinary vehicles, and therefore require greater safety supervision.

[0003] At present, road safety supervision faces problems such as information islands and data fragmentation. Especially for the safety management of some key vehicles, a single data source cannot fully reflect the operating status and potential risks of the vehicle. In order to improve the safety management capabilities of key vehicles, when the number of data sources collected increases, current technology cannot accurately and scientifically analyze multi-source heterogeneous data.

[0004] Therefore, there is currently no standardized model to rationalize and intelligently coordinate the management of vehicle travel data portraits and the risks they face, making it impossible to establish a safer, more efficient and intelligent key vehicle transportation system. Summary of the invention

[0005] In order to improve the safety management capabilities of key vehicles, the present application provides a method and system for constructing travel profiling and risk assessment of key vehicles.

[0006] In the first aspect, the present application provides a method for constructing a travel profile and risk assessment of key vehicles, using the following technical solutions:

[0007] A method for constructing a travel profile and risk assessment of key vehicles includes the following steps:

[0008] Collect multi-source data and pre-process them;

[0009] Integrate the pre-processed multi-source data into a preset unified database;

[0010] fusing the plurality of multi-source data in the unified database to achieve subject classification to obtain a plurality of data subjects, and performing association alignment on the plurality of data subjects to obtain association relationships between the plurality of data subjects;

[0011] Extracting key feature information corresponding to a number of the multi-source data in the unified database and integrating them to construct feature candidate libraries corresponding to a number of different data subjects, the feature candidate libraries including a vehicle feature candidate library, a driver feature candidate library, an enterprise feature candidate library, and a cargo feature candidate library;

[0012] Building a label system based on the candidate feature libraries and the association relationships combined with a portrait evaluation algorithm, and determining the portrait label type corresponding to the key vehicle based on the label system;

[0013] Setting multiple risk assessment dimensions, and constructing a risk assessment score result including multiple risk assessment dimensions based on several of the feature candidate libraries combined with a risk assessment algorithm;

[0014] A travel profile of key vehicles is constructed by combining the portrait label type and the risk assessment score result.

[0015] In some embodiments, collecting multi-source data and preprocessing it includes the following steps:

[0016] Selecting a data source channel to collect the multi-source data, the multi-source data including but not limited to key vehicle basic information, driver basic information, enterprise basic information, and cargo basic information;

[0017] Performing a cleaning process on the multi-source information, the cleaning process comprising:

[0018] When there are missing values ​​in the multi-source data, fill the blank positions corresponding to the missing values ​​by deleting, interpolating or replacing abnormal values;

[0019] Performing consistency check on the multi-source data;

[0020] De-duplication of duplicate data in the multi-source data.

[0021] In some embodiments, integrating the pre-processed multi-source data into a preset unified database comprises the following steps:

[0022] Acquire data characteristic information of some of the multi-source data, the data characteristic information including structure, field format, data type, and meaning, and determine the source and function of the multi-source data according to the data characteristic information;

[0023] Mapping the field formats between the multiple source data to the same field name to filter target data to be processed, wherein the target data to be processed is characterized as data with inconsistent field names or mismatched data types after mapping;

[0024] Performing field conversion on the target data to be processed so that the plurality of multi-source data maintain a unified format;

[0025] A plurality of the multi-source data are aggregated to obtain a unified data set and then stored in the unified database.

[0026] In some of the embodiments, extracting key feature information corresponding to a number of the multi-source data in the unified database and integrating them to construct a feature candidate library corresponding to a number of different data subjects includes the following steps:

[0027] Selecting the data subject according to the data characteristic information, the data subject includes key vehicles, drivers, affiliated companies, and transported goods;

[0028] Extracting basic features and analysis features corresponding to different data subjects from the unified data set based on the sources and functions of the multi-source data, wherein the basic features are represented by original data features directly extracted from the unified data set, and the analysis features are represented by data features obtained after secondary analysis of the basic features;

[0029] The basic features and / or the analysis features are matched to the corresponding data subjects and integrated to construct a plurality of feature candidate libraries.

[0030] In some embodiments, a label system is constructed based on the feature candidate libraries and the association relationships combined with a portrait evaluation algorithm, and the type of portrait label corresponding to the key vehicle is determined based on the label system, including the following steps:

[0031] Building a labeling system based on the portrait evaluation algorithm including the DBSCAN classification and clustering algorithm, vectorized representation, and time series model, wherein the labeling system includes several preset and different types of portrait labels;

[0032] Divide the data subject corresponding to each of the feature candidate libraries into similar clusters based on the DBSCAN classification and clustering algorithm and define them as the first processing result;

[0033] Vectorizing each of the key feature information to obtain the association relationship and potential risk attributes between the multi-source data of the key vehicle and define them as a second processing result;

[0034] Based on a preset time series model, the temporal and spatial risk change information of the key vehicle is determined to predict the future temporal and spatial behavior of the key vehicle to obtain a third processing result;

[0035] The type of portrait label corresponding to the key vehicle is determined in the label system based on the first processing result, the second processing result and the third processing result.

[0036] In some of the embodiments, constructing a risk assessment score result including multiple risk assessment dimensions based on several of the feature candidate libraries combined with a risk assessment algorithm includes the following steps:

[0037] Mapping the key feature information to the interval [0, 1] to perform deviation standardization;

[0038] A risk assessment angle is selected based on the feature candidate library, and the corresponding risk assessment dimension is selected based on the risk assessment angle to calculate the risk assessment score result. The risk assessment angle includes vehicle status, traffic frequency, vehicle accidents, vehicle violations, and cargo type.

[0039] In some embodiments, selecting the corresponding risk assessment dimension based on the risk assessment perspective to calculate the risk assessment score result includes the following steps:

[0040] The risk assessment score is calculated using the following formula:

[0041] ,

[0042] Among them, k is the type coefficient of key vehicles, The status of the key vehicle, is the cargo type, The vehicle traffic frequency, Characterized by the types, number and responsibility of traffic accidents involving key vehicles, Characterize the types, frequency and harm of traffic violations involving key vehicles.

[0043] In some of the embodiments, after building a key vehicle travel portrait in combination with the portrait label type and the risk assessment score result, the following steps are included:

[0044] Generate corresponding entity points based on the types of the portrait tags, and generate relationship lines corresponding to each of the entity points according to the association relationship between the data subjects, where one entity point corresponds to one or more relationship lines;

[0045] Defining the plurality of relationship links as strong relationship links, weak relationship links and regular relationship links based on the risk assessment scores;

[0046] The entity map corresponding to the key vehicle is generated through the entity points and the relationship lines.

[0047] In some of the embodiments, the following steps are also included:

[0048] Generate a key vehicle management framework, the key vehicle management framework includes a data layer, a feature extraction layer, an algorithm layer, an indicator layer and an application layer, wherein:

[0049] The data layer is used to add a number of the multi-source data, and divide the multi-source data into dynamic data, static data and basic data;

[0050] The indicator layer is used to add some of the key feature information;

[0051] The algorithm layer is used to add the algorithms needed to be used in the process of constructing the key vehicle travel portrait and the risk assessment score result;

[0052] The indicator layer is used to add the corresponding key vehicle travel portraits, and divide the key vehicle travel portraits into key vehicle portraits, driver portraits and enterprise portraits;

[0053] The application layer is used to add application functions required based on the travel profile of the key vehicles.

[0054] In the second aspect, the present application provides a key vehicle travel profiling and risk assessment construction system, which adopts the following technical solutions:

[0055] A key vehicle travel profiling and risk assessment construction system, including:

[0056] The acquisition and processing module is used to acquire multi-source data and perform pre-processing;

[0057] A fusion processing module is used to integrate the pre-processed multi-source data into a preset unified database; and fuse the multi-source data in the unified database to achieve subject classification to obtain a number of data subjects, and associate and align the data subjects to obtain the association relationship between the data subjects;

[0058] A feature candidate library building module is used to extract key feature information corresponding to a number of the multi-source data in the unified database and integrate them to build a number of feature candidate libraries corresponding to different data subjects, wherein the feature candidate libraries include a vehicle feature candidate library, a driver feature candidate library, an enterprise feature candidate library, and a cargo feature candidate library;

[0059] A key vehicle travel portrait label generation module, which constructs a label system based on the feature candidate libraries and the association relationship combined with a portrait evaluation algorithm, and determines the type of portrait label corresponding to the key vehicle based on the label system;

[0060] A risk assessment module sets a plurality of risk assessment dimensions and constructs a risk assessment score result including the plurality of risk assessment dimensions based on a plurality of the feature candidate libraries combined with a risk assessment algorithm;

[0061] The portrait construction module is used to construct a travel portrait of key vehicles by combining the portrait label type and the risk assessment score result.

[0062] Through the above technical solution, this application has the following technical effects:

[0063] By comprehensively utilizing various multi-source data of key vehicles and adopting corresponding algorithm technology, a label system including personnel, vehicles, cargo, enterprises, etc. is constructed, and travel profiles and risk assessments of key vehicles are constructed based on this label system. The travel profiles of key vehicles are used to deeply mine and analyze user data, and the potential risks of key vehicles are predicted through a reasonable risk assessment system, thereby effectively improving the monitoring and management efficiency of key vehicles and providing decision-making support for traffic management departments. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a schematic diagram of the steps of the method for constructing a key vehicle travel profile and risk assessment provided in an embodiment of the present application.

[0065] Figure 2 It is a schematic diagram of the framework of the key vehicle management architecture in the embodiment of the present application.

[0066] Figure 3 It is a module connection diagram of the key vehicle travel profiling and risk assessment construction system in the embodiment of this application. DETAILED DESCRIPTION

[0067] To more clearly understand the purpose, technical solutions and advantages of the present application, the present application is described and illustrated below in conjunction with the accompanying drawings and embodiments. However, it should be understood by those of ordinary skill in the art that the present application can be implemented without these details. In some cases, in order to avoid unnecessary descriptions that make various aspects of the present application obscure, well-known methods, processes, systems, components and / or circuits that have been described at a higher level will not be described in detail. For those of ordinary skill in the art, it is obvious that various changes can be made to the embodiments disclosed in the present application, and without departing from the principles and scope of the present application, the general principles defined in the present application can be applied to other embodiments and application scenarios. Therefore, the present application is not limited to the embodiments shown, but conforms to the broadest scope consistent with the scope claimed for protection of the present application.

[0068] In the description of this application, "several" means one or more, "more" means more than two, "greater than", "less than", "exceed", etc. are understood to exclude the number itself, and "above", "below", "within", etc. are understood to include the number itself. If there is a description of "first" or "second", it is only used to distinguish the technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0069] In the description of the present application, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples.

[0070] refer to Figures 1 to 3 , an embodiment of the present application discloses a method for constructing a travel profile and risk assessment of key vehicles.

[0071] like Figure 1 As shown, a method for constructing a travel profile and risk assessment of key vehicles includes the following steps:

[0072] S100, collecting multi-source data and performing pre-processing.

[0073] First, the reference theoretical data required for portrait construction and risk assessment are collected. The data can be obtained from various channels through the Internet of Things data. After obtaining multi-source data, in order to facilitate the subsequent corresponding data processing efficiency and accuracy, the collected multi-source data needs to be preprocessed.

[0074] S200, integrating the pre-processed multi-source data into a preset unified database.

[0075] The processed multi-source data is unified so that several data with different structures are changed into the same format and structure to facilitate subsequent processing.

[0076] S300, a plurality of multi-source data in a unified database are merged to achieve subject classification to obtain a plurality of data subjects, and the plurality of data subjects are associated and aligned to obtain the association relationship between the plurality of data subjects.

[0077] By analyzing the unified multi-source data to extract the data content corresponding to each data, and classifying the multi-source data according to the data content, several different data subjects are divided based on the preset requirements during classification. The data content is analyzed to obtain the association relationship between the data subjects, which is characterized by the influence of one data subject on another data subject, the association effect between each other, and the combined influence.

[0078] S400, extracting key feature information corresponding to a number of multi-source data in a unified database and integrating them to construct a feature candidate library corresponding to a number of different data subjects.

[0079] Data content determination and extraction are performed on each multi-source data to obtain key feature information corresponding to the multi-source data. The key feature information is used to represent the attributes, functions and objects corresponding to a data.

[0080] After the key feature information is extracted, it is grouped and integrated according to its corresponding content information to obtain several feature candidate libraries corresponding to different data subjects, and each feature candidate library stores several key feature information used to represent the attributes of the data subject.

[0081] The feature candidate library includes a vehicle feature candidate library, a driver feature candidate library, an enterprise feature candidate library, and a cargo feature candidate library.

[0082] S500, constructing a label system based on several feature candidate libraries and association relationships combined with a portrait assessment algorithm, and determining the type of portrait label corresponding to the key vehicle based on the label system.

[0083] Through the preset portrait assessment algorithm and the association relationship between each data subject, a label system is constructed based on key feature information in the feature candidate library. This label system is used to conduct portrait evaluation around key vehicles. After evaluating the vehicles, people, enterprises, goods and other contents in the key vehicles based on multi-source data, the portrait label corresponding to the key vehicle can be obtained.

[0084] S600, setting multiple risk assessment dimensions, and constructing a risk assessment score result including multiple risk assessment dimensions based on a plurality of feature candidate libraries combined with a risk assessment algorithm.

[0085] A risk assessment model is constructed by combining the risk assessment algorithm with the key feature information in the feature candidate library. The risk assessment model corresponds to several risk assessment dimensions. The risk assessment score results corresponding to key vehicles are obtained through multi-dimensional risk assessment.

[0086] S700, combines the portrait label types and risk assessment score results to build travel portraits of key vehicles.

[0087] Once the portrait label type and risk assessment score results are obtained, the two can be combined to obtain a travel portrait of key vehicles that is used to evaluate the operating status and safety level of key vehicles.

[0088] Through the above-mentioned technical solution, various multi-source data of key vehicles are comprehensively utilized and corresponding algorithm technology is adopted to construct a label system including personnel, vehicles, goods, enterprises, etc., and based on this label system, travel portraits and risk assessments of key vehicles are constructed. The travel portraits of key vehicles are used to deeply mine and analyze user data, and the potential risks of key vehicles are predicted through a reasonable risk assessment system, thereby effectively improving the monitoring and management efficiency of key vehicles and providing decision-making support for traffic management departments.

[0089] In some other embodiments, collecting multi-source data and preprocessing it includes the following steps:

[0090] S110, selecting a data source channel to collect multi-source data, where the multi-source data includes but is not limited to basic information of key vehicles, basic information of drivers, basic information of enterprises, and basic information of goods.

[0091] It can also include checkpoint monitoring data, radar monitoring data, video surveillance image data, etc.

[0092] S120, performing cleaning processing on the multi-source information, the cleaning processing includes: when there are missing values ​​in the multi-source data, filling the blank positions corresponding to the missing values ​​by deleting, interpolating or replacing abnormal values; performing consistency check on the multi-source data; and deduplicating duplicate data in the multi-source data.

[0093] Preprocessing is to clean a large amount of data. When there is missing data, the missing blanks are filled by deleting, interpolating (mean, median) or replacing abnormal values. If abnormal data is detected, the abnormal data is processed to ensure the rationality of the data. Consistency checks are performed on some data, such as checking dates and value ranges to ensure the logical coherence of the data. Data deduplication is to remove duplicate content from multiple multi-source data.

[0094] In some other embodiments, integrating the pre-processed multi-source data into a preset unified database includes the following steps:

[0095] S210, obtaining data characteristic information of a plurality of multi-source data, and determining the source and function of the multi-source data according to the data characteristic information.

[0096] Data characteristic information includes structure, field format, data type, and meaning.

[0097] Fully understand the data characteristics of each multi-source data, and clarify the source of the multi-source data and its role in the overall analysis based on its data characteristics.

[0098] S220, mapping the field formats between the various multi-source data to the same field name to filter the target data to be processed.

[0099] The target data to be processed is characterized by data with inconsistent field names or mismatched data types after mapping.

[0100] S230, performing field conversion on the target data to be processed so that the plurality of multi-source data maintain a unified format.

[0101] For different fields of different multi-source data, the relationship between the fields is first clarified through mapping. Specifically, similar fields in different multi-source data are mapped to common field names, and unit conversion is performed to ensure consistency of measurement units and timeline synchronization, so that multiple multi-source data remain consistent.

[0102] S240, a plurality of multi-source data are collected to obtain a unified data set and the unified data set is merged and stored in a unified database.

[0103] The multi-source data that has been cleaned and formatted is collected and integrated into a unified data set and stored in a unified database. At the same time, it is also necessary to detect and resolve conflicts in the integrated data set based on the unique identifier of the data to avoid duplicate data in the data set.

[0104] In some other embodiments, extracting key feature information corresponding to a plurality of multi-source data in a unified database and integrating them to construct a feature candidate library corresponding to a plurality of different data subjects includes the following steps:

[0105] S410, selecting a data subject according to the data characteristic information.

[0106] The data subjects include key vehicles, drivers, affiliated companies, and transported goods.

[0107] S420, extracting basic features and analysis features corresponding to different data subjects from a unified data set based on the sources and functions of multi-source data.

[0108] The basic feature representation is the original data feature extracted directly from the unified data set without any processing or conversion. The judgment feature representation is the new data feature obtained after secondary analysis of the basic features. The secondary analysis includes the calculation of a certain amount of data through frequency and time.

[0109] Specifically,

[0110] For key vehicles, the basic characteristics include vehicle type, nature of use, vehicle load, vehicle age, insurance, whether inspection is overdue, whether scrapping is overdue, owner, company, vehicle registration place, etc. The analysis and judgment characteristics include frequency of passage, frequency of violations, frequency of accidents, number of fatigue driving, average mileage, average driving time, and distribution of driving areas.

[0111] For drivers, basic characteristics include age, gender, driver's license status, operating qualifications, points, etc., and analysis characteristics include driving experience, accident frequency, violation frequency, etc.

[0112] For the affiliated companies, the basic characteristics include main business, main operating area, company age, etc. The analysis and judgment characteristics include the total number of vehicles, accident frequency, violation frequency, industry experience, vehicle utilization rate, average mileage, etc.

[0113] For goods, it only includes basic characteristics, such as the type of goods, degree of danger, quality of goods, etc.

[0114] It should be noted that when constructing the subsequent feature candidate library, depending on the application scenario and target requirements, it can be constructed through a combination of basic features and analysis features, or only through basic features. At the same time, the missing of some features that cannot be collected is allowed. When some key feature information is missing, the missing values ​​can be analyzed through deletion, filling, prediction, etc. to reduce the impact on subsequent calculations and evaluations.

[0115] S430, the basic features and / or the analysis and judgment features are matched to the corresponding data subjects and integrated to construct a plurality of feature candidate libraries.

[0116] In other embodiments, a label system is constructed based on several feature candidate libraries and association relationships combined with a portrait evaluation algorithm, and the type of portrait label corresponding to the key vehicle is determined based on the label system, including the following steps:

[0117] S510, building a labeling system based on a portrait assessment algorithm including a DBSCAN classification and clustering algorithm, a vectorized representation, and a time series model, wherein the labeling system includes a number of preset and different types of portrait labels.

[0118] In combination with the above-mentioned feature candidate library, a label system is established through a multi-level and multi-class classification rule engine. In the embodiment of the present application, the multi-level and multi-class classification rule engine is respectively composed of a DBSCAN classification clustering algorithm, a vectorized representation, and a rule algorithm corresponding to the time series model.

[0119] S520, dividing the data subjects corresponding to each feature candidate library into similar clusters based on the DBSCAN classification and clustering algorithm and defining them as the first processing result.

[0120] First, we use the DBSCAN classification and clustering algorithm to divide vehicles, people, enterprises, and goods into similar clusters. Specifically,

[0121] Initialization parameters: Setting the neighborhood radius (epsilon), used to find points around a core point; sets the minimum number of points that need to exist in the neighborhood of a point , so that the point is considered a core point.

[0122] Data point classification: Initialize each point in the dataset to an unvisited state.

[0123] For each point in the dataset, do the following:

[0124] If the point has not been visited, mark it as visited;

[0125] Count the number of points in the neighborhood of the point (at distance points within (epsilon);

[0126] If the neighborhood of a point contains at least Points are marked as core points;

[0127] Otherwise, if it is within the neighborhood of other core points, it is marked as a boundary point;

[0128] If it is neither a core point nor part of the neighborhood of any core point, it is marked as a noise point.

[0129] Construct clusters:

[0130] Select a core point from the unassigned points as a seed point;

[0131] Recursively find all The core points within the (epsilon) radius are added to the same cluster;

[0132] For each newly added point, if it is also a core point, continue to recursively search for points in its neighborhood until there are no more points to add;

[0133] Assign all border points to the cluster to which their associated core point belongs.

[0134] Finally, repeat the above steps until all data points are processed, all points are either marked as noise points or assigned to a cluster, and finally the set of all clusters and possible noise points are output to complete the classification of all key feature information.

[0135] S530, vectorizing each key feature information to obtain the correlation relationship and potential risk attributes between the multi-source data of the key vehicle and define them as the second processing result.

[0136] Based on the vectorization representation of key feature information, a deep neural network is used to capture the complex nonlinear relationship and potential risk attributes between vehicle behaviors. Feature vectorization can use one-hot encoding. For example, if you want to capture the relationship between illegal behavior and fatigue driving, you can include the following steps:

[0137] First, determine the categorical variables. Suppose there is a feature called "violation", which contains six categories: "illegal_parking", "driving_in_the_wrong_lane", "illegal_lane_change", "drive_and_make_phone_calls", "low_speed", and "over_speed".

[0138] Second, dummy variables are created. For each unique categorical value, a new binary column (dummy or dummy variable) is created. For example, the "violation" feature is converted into six new columns: "violation_illegal_parking", "violation_driving_in_the_wrong_lane", "violation_illegal_lane_change", "violation_drive_and_make_phone_calls", "violation_low_speed", "violation_over_speed".

[0139] Each row of the original data is encoded. If the "violation" feature value of the row is equal to a certain category (for example, "illegal_parking"), the corresponding dummy variable "violation_illegal_parking" is encoded as 1, and the other dummy variables are 0. If the "violation" feature value does not belong to any category, all dummy variables will be encoded as 0.

[0140] Choose to replace the original categorical variable column or add the generated dummy variable column to the existing data set to avoid redundant information in the data set.

[0141] If there are missing values ​​in the categorical variable, you need to create an additional dummy variable for the missing values ​​during one-hot encoding.

[0142] When a categorical variable has k different values, theoretically only k-1 dummy variables are needed to fully represent this variable, because the kth category can be implicitly represented by the situation where all other categories are 0. In order to avoid multicollinearity, one dummy variable is usually discarded and only k-1 are retained.

[0143] In the same way, the fatigue driving label "fatigue_driving" is feature encoded, with "violation" as input and "fatigue_driving" as output. The deep neural network is used to learn the complex nonlinear relationship between the two, so as to judge the probability of fatigue driving behavior of key vehicles based on the violation.

[0144] S540, based on the preset time series model, determine the time-space risk change information of the key vehicle to predict the future time-space behavior of the key vehicle to obtain a third processing result.

[0145] Based on the LSTM time series model, the risk changes of key vehicles in the time and space domain are captured, for example, the future illegal behaviors of key vehicles are predicted. The calculation method of the LSTM model is as follows:

[0146] ;

[0147] ;

[0148] ;

[0149] ;

[0150] ;

[0151] .

[0152] in, Characterized as input gate, Characterized by the forget gate, Characterized as output gate, Characterized by cell state updates, Characterized by cell state, Represented as the cell state at the previous moment, Represented as the current hidden state, Represented as the hidden state of the previous moment, for Activation function, is the hyperbolic tangent activation function, Represented as the input gate weight matrix, Represented as the candidate memory cell weight matrix, Represented as the output gate weight matrix, is the forget gate bias term, Characterized as the input gate bias term, Characterized as the output gate bias term, Characterized as candidate memory cell bias term, Enter for the current moment.

[0153] S550: Determine the type of portrait label corresponding to the key vehicle in the label system based on the first processing result, the second processing result, and the third processing result.

[0154] After the three corresponding processing results are calculated according to the three algorithms, the type of portrait label corresponding to the key vehicle can be determined in the label system. This type is used to indicate the data subject corresponding to the key vehicle, the corresponding state type, its own safety hazards, etc.

[0155] In other embodiments, constructing a risk assessment score result including multiple risk assessment dimensions based on several feature candidate libraries combined with a risk assessment algorithm includes the following steps:

[0156] S610, mapping some key feature information to the interval [0, 1] to perform deviation standardization.

[0157] The key feature information is normalized to eliminate the impact of data unit magnitude and large or small data differences. The calculation method is as follows:

[0158] ,

[0159] in, Represents newly generated data. Represents the original data, Refers to the minimum value in the original data. Refers to the maximum value in the original data.

[0160] S620, selecting a risk assessment angle based on the feature candidate library, and selecting a corresponding risk assessment dimension based on the risk assessment angle to calculate a risk assessment score result.

[0161] Among them, risk assessment angles include vehicle status, traffic frequency, vehicle accidents, vehicle violations, and cargo type.

[0162] The risk assessment dimensions include the status of key vehicles, cargo type, vehicle traffic frequency, type, number, and responsibility of traffic accidents involving vehicles, and type, number, and hazards of traffic violations involving vehicles.

[0163] In other embodiments, selecting corresponding risk assessment dimensions based on the risk assessment perspective to calculate the risk assessment score result includes the following steps:

[0164] S621, the risk assessment score is calculated using the following formula:

[0165] ,

[0166] Among them, k is the type coefficient of key vehicles, The status of the key vehicle, is the cargo type, The vehicle traffic frequency, Characterized by the types, number and responsibility of traffic accidents involving key vehicles, Characterize the types, frequency and harm of traffic violations involving key vehicles.

[0167] Through the above calculation formula, by adding multi-source data with corresponding key feature information, we can obtain the risk assessment score results corresponding to key vehicles covering different angles and dimensions.

[0168] In the embodiment of the present application, different risk assessment score results correspond to different risk levels, 0-30 points represent low risk, corresponding to the blue color prompt, 30-65 points represent relatively low risk, corresponding to the green color prompt, 65-80 points represent medium risk, corresponding to the yellow color prompt, 80-90 points represent relatively high risk, corresponding to the orange color prompt, 90-100 points represent high risk, corresponding to the red color prompt.

[0169] In some other embodiments, after building a key vehicle travel portrait by combining the portrait label type and the risk assessment score result, the following steps are also included:

[0170] S800, generating corresponding entity points based on the type of portrait label, generating relationship lines corresponding to each entity point according to the association relationship between data subjects, and one entity point corresponds to one or more relationship lines.

[0171] Generate entity points corresponding to each multi-source data, and the entity points are connected to each other through relationship lines corresponding to the association relationship. According to different association relationships between multi-source data, each entity point corresponds to one or more relationship lines.

[0172] S810, defining a number of relationship links as strong relationship links, weak relationship links, and regular links based on risk assessment scores.

[0173] A strong relationship line indicates that one multi-source data has a strong influence on another multi-source data, such as the violation frequency of key vehicles and the violation frequency of the total vehicles of the company to which it belongs. It means that the change of one data is strongly correlated to the change of another data.

[0174] Weak relationship links indicate that one multi-source data has a weak influence on another multi-source data, such as the vehicle registration location and the driver's gender of a key vehicle, which means that the relationship of one data has a weak correlation with the relationship of another data.

[0175] Conventional connections represent that the influence relationship between one multi-source data and several multi-source data are all conventional relationships.

[0176] S820, generating a physical map corresponding to the key vehicles through physical points and relationship lines.

[0177] Through relationship lines, each entity point is connected to obtain an overall entity map, and the key vehicle labels are structured and expressed in turn. The relationship between key vehicles is deeply mined and displayed, enriching the travel portrait content of key vehicles.

[0178] like Figure 2 As shown, in some other embodiments, the following steps are also included:

[0179] S900, generates a key vehicle management architecture, which includes data layer, feature extraction layer, algorithm layer, indicator layer and application layer.

[0180] S910, the data layer is used to add a number of multi-source data, and divide the multi-source data into dynamic data, static data and basic data.

[0181] Dynamic data includes checkpoint data, ETC data, video surveillance data, and radar data; static data includes highway networks and urban road networks; basic data includes vehicle basic data, driver basic data, enterprise basic data, and cargo basic data.

[0182] S920, the indicator layer is used to add some key feature information.

[0183] The indicator layer includes five categories:

[0184] Vehicle attributes, including: vehicle type, model, size and load, maintenance status, technical level, and age.

[0185] Operational information, including: company, transportation routes, and freight categories.

[0186] Driving behavior characteristics include: speeding percentage, average speed, violation frequency, accident frequency, frequently driven expressways, and number of instances of fatigue driving.

[0187] Operation status, including: operation cost, whether or not an ETC user, and operation time preference.

[0188] Risk assessment, including: violation and accident rates, transportation compliance, overloading and over-limit situations.

[0189] S930, the algorithm layer is used to add the algorithms needed in the process of constructing the travel portrait of key vehicles and the risk assessment score structure.

[0190] The algorithm layer includes knowledge graphs, DBSCAN classification and clustering algorithms, vectorized representations, time series models, and risk assessment algorithms.

[0191] S940, the indicator layer is used to add corresponding key vehicle travel portraits, and divide the key vehicle travel portraits into key vehicle portraits, driver portraits and enterprise portraits.

[0192] Among them, key vehicle portraits include vehicle status, vehicles with frequent violations, vehicles with frequent accidents, and high-risk vehicles; driver portraits include driver's license status, people with frequent violations, people with frequent accidents, and credit evaluation; enterprise portraits include enterprise type, enterprise scale, enterprise hidden dangers, and enterprise risks.

[0193] S950, the application layer is used to add application functions based on the travel profile requirements of key vehicles.

[0194] The application layer includes functional applications that can be achieved through various portraits and risk assessment results, including key vehicle monitoring, risk identification and prediction, proactive control, real-time intervention, and integrated management of people, vehicles and companies.

[0195] Through the above method, a complete vehicle management architecture is built, through which various types of information are integrated, and after integration, they are connected to the corresponding application systems to realize different vehicle management functions.

[0196] like Figure 3 As shown, the present application also discloses a key vehicle travel portrait and risk assessment construction system, which includes:

[0197] The acquisition and processing module is used to acquire multi-source data and perform preprocessing.

[0198] The fusion processing module is used to integrate the pre-processed multi-source data into a preset unified database; and fuse the multi-source data in the unified database to achieve subject classification to obtain a number of data subjects, and associate and align the data subjects to obtain the association relationship between the data subjects.

[0199] The feature candidate library building module is used to extract key feature information corresponding to several multi-source data in the unified database and integrate them to build feature candidate libraries corresponding to several different data subjects. The feature candidate libraries include vehicle feature candidate libraries, driver feature candidate libraries, enterprise feature candidate libraries, and cargo feature candidate libraries.

[0200] The key vehicle travel portrait label generation module builds a label system based on several feature candidate libraries and association relationships combined with a portrait assessment algorithm, and determines the type of portrait label corresponding to the key vehicles based on the label system.

[0201] The risk assessment module sets multiple risk assessment dimensions and constructs a risk assessment score result containing multiple risk assessment dimensions based on several feature candidate libraries combined with risk assessment algorithms.

[0202] The portrait construction module is used to construct travel portraits of key vehicles by combining the portrait label types and risk assessment score results.

[0203] The implementation principle is:

[0204] By comprehensively utilizing various multi-source data of key vehicles and adopting corresponding algorithm technology, a label system including personnel, vehicles, cargo, enterprises, etc. is constructed, and travel profiles and risk assessments of key vehicles are constructed based on this label system. The travel profiles of key vehicles are used to deeply mine and analyze user data, and the potential risks of key vehicles are predicted through a reasonable risk assessment system, thereby effectively improving the monitoring and management efficiency of key vehicles and providing decision-making support for traffic management departments.

[0205] It should be understood that although the steps in the flowchart of the accompanying drawings are shown in sequence according to the instructions of the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise clearly stated in this document, the execution of these steps is not strictly limited in order and can be performed in other orders.

[0206] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for constructing a travel profile and risk assessment of key vehicles, characterized in that: The following steps are involved: Collect multi-source data and pre-process them; Integrate the pre-processed multi-source data into a preset unified database; fusing the plurality of multi-source data in the unified database to achieve subject classification to obtain a plurality of data subjects, and performing association alignment on the plurality of data subjects to obtain association relationships between the plurality of data subjects; Extracting key feature information corresponding to a number of the multi-source data in the unified database and integrating them to construct feature candidate libraries corresponding to a number of different data subjects, the feature candidate libraries including a vehicle feature candidate library, a driver feature candidate library, an enterprise feature candidate library, and a cargo feature candidate library; A label system is constructed based on the candidate feature libraries and the association relationships combined with a portrait evaluation algorithm, and the portrait label type corresponding to the key vehicle is determined based on the label system. Specifically, Building a labeling system based on the portrait evaluation algorithm including the DBSCAN classification and clustering algorithm, vectorized representation, and time series model, wherein the labeling system includes several preset and different types of portrait labels; Divide the data subject corresponding to each of the feature candidate libraries into similar clusters based on the DBSCAN classification and clustering algorithm and define them as the first processing result; Vectorizing each of the key feature information to obtain the association relationship and potential risk attributes between the multi-source data of the key vehicle and define them as a second processing result; Based on a preset time series model, the temporal and spatial risk change information of the key vehicle is determined to predict the future temporal and spatial behavior of the key vehicle to obtain a third processing result; Determining the type of portrait label corresponding to the key vehicle in the label system based on the first processing result, the second processing result, and the third processing result; Set multiple risk assessment dimensions, and construct a risk assessment score result containing multiple risk assessment dimensions based on several of the feature candidate libraries combined with the risk assessment algorithm. Specifically, Mapping the key feature information to the interval [0, 1] to perform deviation standardization; A risk assessment angle is selected based on the feature candidate library, and a corresponding risk assessment dimension is selected based on the risk assessment angle to calculate a risk assessment score result, wherein the risk assessment angle includes vehicle status, traffic frequency, vehicle accidents, vehicle violations, and cargo type; The risk assessment score is calculated using the following formula: , Among them, k is the type coefficient of key vehicles, The status of the key vehicle, is the cargo type, The vehicle traffic frequency, Characterized by the types, number and responsibility of traffic accidents involving key vehicles, Characterize the types, number and harm of traffic violations involving key vehicles; A travel profile of key vehicles is constructed by combining the portrait label type and the risk assessment score result.

2. The method for constructing a travel profile and risk assessment of key vehicles according to claim 1 is characterized in that: Collecting multi-source data and preprocessing it includes the following steps: Selecting a data source channel to collect the multi-source data, the multi-source data including but not limited to key vehicle basic information, driver basic information, enterprise basic information, and cargo basic information; The multi-source data is cleansed, and the cleaning process includes: When there are missing values ​​in the multi-source data, fill the blank positions corresponding to the missing values ​​by deleting, interpolating or replacing abnormal values; Performing consistency check on the multi-source data; De-duplication of duplicate data in the multi-source data.

3. The method for constructing key vehicle travel profile and risk assessment according to claim 1 is characterized in that: Integrating the pre-processed multi-source data into a preset unified database comprises the following steps: Acquire data characteristic information of some of the multi-source data, the data characteristic information including structure, field format, data type, and meaning, and determine the source and function of the multi-source data according to the data characteristic information; Mapping the field formats between the multiple source data to the same field name to filter target data to be processed, wherein the target data to be processed is characterized as data with inconsistent field names or mismatched data types after mapping; Performing field conversion on the target data to be processed so that the multi-source data maintain a unified format; A plurality of the multi-source data are aggregated to obtain a unified data set and stored in the unified database.

4. The method for constructing a travel profile and risk assessment of key vehicles according to claim 3 is characterized in that: Extracting key feature information corresponding to a number of the multi-source data in the unified database and integrating them to construct a feature candidate library corresponding to a number of different data subjects, including the following steps: Selecting the data subject according to the data characteristic information, the data subject includes key vehicles, drivers, affiliated companies, and transported goods; Extracting basic features and analysis features corresponding to different data subjects from the unified data set based on the sources and functions of the multi-source data, wherein the basic features are represented by original data features directly extracted from the unified data set, and the analysis features are represented by data features obtained after secondary analysis of the basic features; The basic features and / or the analysis features are matched to the corresponding data subjects and integrated to construct a plurality of feature candidate libraries.

5. The method for constructing key vehicle travel profile and risk assessment according to claim 1 is characterized in that: After constructing the key vehicle travel portrait in combination with the portrait label type and the risk assessment score result, the following steps are included: Generate corresponding entity points based on the types of the portrait tags, and generate relationship lines corresponding to each of the entity points according to the association relationship between the data subjects, where one entity point corresponds to one or more relationship lines; Defining the plurality of relationship links as strong relationship links, weak relationship links and regular relationship links based on the risk assessment scores; The entity map corresponding to the key vehicle is generated through the entity points and the relationship lines.

6. The method for constructing a travel profile and risk assessment of key vehicles according to claim 5 is characterized in that: The following steps are also included: Generate a key vehicle management framework, the key vehicle management framework includes a data layer, a feature extraction layer, an algorithm layer, an indicator layer and an application layer, wherein: The data layer is used to add a number of the multi-source data, and divide the multi-source data into dynamic data, static data and basic data; The indicator layer is used to add some of the key feature information; The algorithm layer is used to add the algorithms needed to be used in the process of constructing the key vehicle travel portrait and the risk assessment score result; The indicator layer is used to add the corresponding key vehicle travel portraits, and divide the key vehicle travel portraits into key vehicle portraits, driver portraits and enterprise portraits; The application layer is used to add application functions required based on the travel profile of the key vehicles.

7. A key vehicle travel profile and risk assessment construction system, characterized in that: include: The acquisition and processing module is used to acquire multi-source data and perform pre-processing; A fusion processing module, used for integrating the pre-processed multi-source data into a preset unified database; and fusing the plurality of multi-source data in the unified database to achieve subject classification to obtain a plurality of data subjects, and performing association alignment on the plurality of data subjects to obtain association relationships between the plurality of data subjects; A feature candidate library building module is used to extract key feature information corresponding to a number of the multi-source data in the unified database and integrate them to build a number of feature candidate libraries corresponding to different data subjects, wherein the feature candidate libraries include a vehicle feature candidate library, a driver feature candidate library, an enterprise feature candidate library, and a cargo feature candidate library; The key vehicle travel portrait label generation module constructs a label system based on the feature candidate libraries and the association relationship combined with the portrait evaluation algorithm, and determines the type of portrait label corresponding to the key vehicle based on the label system. Specifically, Building a labeling system based on the portrait evaluation algorithm including the DBSCAN classification and clustering algorithm, vectorized representation, and time series model, wherein the labeling system includes several preset and different types of portrait labels; Divide the data subject corresponding to each of the feature candidate libraries into similar clusters based on the DBSCAN classification and clustering algorithm and define them as the first processing result; Vectorizing each of the key feature information to obtain the association relationship and potential risk attributes between the multi-source data of the key vehicle and define them as a second processing result; Based on a preset time series model, the temporal and spatial risk change information of the key vehicle is determined to predict the future temporal and spatial behavior of the key vehicle to obtain a third processing result; Determining the type of portrait label corresponding to the key vehicle in the label system based on the first processing result, the second processing result, and the third processing result; The risk assessment module sets multiple risk assessment dimensions, and constructs a risk assessment score result including multiple risk assessment dimensions based on several of the feature candidate libraries combined with the risk assessment algorithm. Specifically, Mapping the key feature information to the interval [0, 1] to perform deviation standardization; A risk assessment angle is selected based on the feature candidate library, and a corresponding risk assessment dimension is selected based on the risk assessment angle to calculate a risk assessment score result, wherein the risk assessment angle includes vehicle status, traffic frequency, vehicle accidents, vehicle violations, and cargo type; The risk assessment score is calculated using the following formula: , Among them, k is the type coefficient of key vehicles, The status of the key vehicle, is the cargo type, The vehicle traffic frequency, Characterized by the types, number and responsibility of traffic accidents involving key vehicles, Characterize the types, number and harm of traffic violations involving key vehicles; The portrait construction module is used to construct a travel portrait of key vehicles by combining the portrait label type and the risk assessment score result.

Images