Vehicle portrait construction method based on operation data

Abstract

The invention discloses a vehicle portrait construction method based on operation data. The method comprises the following steps: S1, data collection; s2, data cleaning; s3, data integration: classifying the label data cleaned in the step S2 according to the structured degree, static data and dynamic data categories of the label data to form static structured data, static semi-structured data and static non-structured data; the method comprises the following steps: cleaning static semi-structured data, forming dynamic structured data, dynamic semi-structured data and dynamic unstructured data, converting the static semi-structured data into static structured data, converting the dynamic semi-structured data into dynamic structured data, and integrating the cleaned and converted data; s4, creating a database; and S5, according to vehicle portrait construction requirements, constructing a static vehicle portrait or a dynamic vehicle portrait or constructing a total vehicle portrait by combining the static data and the dynamic data. According to the method, the problems of incomplete vehicle attribute definition, low vehicle label construction efficiency, inaccurate vehicle feature representation and the like can be effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle networking, and specifically to a method for constructing a vehicle portrait based on operation data. Background Art

[0002] With the rapid development of the Internet and big data technology, vehicle data is increasing day by day. How to effectively process these massive vehicle data resources and improve the efficiency of vehicle management and analysis. Especially driven by the Internet of Vehicles technology, the collection and analysis of vehicle data has become a common phenomenon. How to make full use of these huge vehicle data to achieve in-depth insights and accurate portraits of vehicle behavior has become the focus of current research.

[0003] Existing vehicle profiling technology still faces many challenges. First, low data processing efficiency is a prominent problem. When faced with large-scale Internet of Vehicles data, existing technologies often require large amounts of computation, take a long time, and have incomplete classification, making it impossible to fully and carefully characterize and analyze vehicle status in real time. The generated vehicle portrait is not complete, and it is difficult to accurately and clearly depict the characteristics of the vehicle.

[0004] When classifying vehicles, the classification results are often not accurate enough due to the incompleteness and ambiguity of the portraits. The bias and incompleteness of the data seriously affect the accuracy and completeness of the vehicle portraits, and thus affect the accuracy of vehicle classification. Traditional vehicle data processing methods often lack effective distinction between data types, resulting in problems such as inefficiency and lack of accuracy when processing large-scale vehicle data. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a vehicle portrait construction method based on operational data, which can effectively solve the problems of incomplete vehicle attribute definition, inefficient vehicle label construction and inaccurate vehicle feature representation in the prior art.

[0006] A vehicle portrait construction method based on operation data includes the following steps:

[0007] S1. Data collection: Collect various original operation data generated by the vehicle during operation;

[0008] S2. Data cleaning: delete unnecessary empty records in step S1, handle abnormal values, fill missing data, remove field duplication, handle error values, unify data formats, and standardize processing;

[0009] S3. Data integration: classify the cleaned label data in step S2 according to their structured degree, static data and dynamic data categories to form static structured data, static semi-structured data and static unstructured data; form dynamic structured data, dynamic semi-structured data and dynamic unstructured data, convert static semi-structured data into static structured data, convert dynamic semi-structured data into dynamic structured data, and integrate the cleaned and converted data;

[0010] S4. Create a database: add static data category information to the static database, add the above dynamic data category information to the dynamic database, add new tags to the database when receiving a tag building task, and update existing tags to the database when receiving a tag update task;

[0011] S5. According to the requirements for constructing vehicle portraits, construct a static vehicle portrait or a dynamic vehicle portrait or combine static data and dynamic data to construct a total vehicle portrait.

[0012] Preferably, the static data is further subdivided into: basic vehicle information, vehicle technical specifications, vehicle dimensions and weight, and vehicle static unstructured data; wherein:

[0013] Basic vehicle information: license plate, model, brand, color, and production date;

[0014] Vehicle technical specifications: fuel vehicle engine information, electric vehicle battery information, electric vehicle motor information, chassis and suspension, tires and brakes;

[0015] Vehicle size and weight: vehicle length, width, height, curb weight, and maximum load;

[0016] Vehicle static unstructured data: image data, video data, audio data, documents, and reports.

[0017] Preferably, the dynamic data is further subdivided into: vehicle trajectory point information, vehicle trajectory segment information, vehicle usage information and vehicle dynamic unstructured data;

[0018] Vehicle trajectory point information: location information, start and end time, start and end time of dangerous driving, total number of speeding alarms, number of sharp turns, number of sudden brakes, and number of sudden accelerations;

[0019] Track segment information: mileage, running time, speeding alarm mileage, maximum speeding alarm duration, fatigue driving mileage, maximum fatigue driving duration, number of fatigue driving times;

[0020] Vehicle usage information: fuel efficiency of fuel vehicles, battery performance of electric vehicles, parking location collection, vehicle fault and diagnostic information;

[0021] Vehicle dynamic unstructured data: log data, video data, audio data.

[0022] Preferably, in step S4, the operational data information that needs to be updated, deleted, or added is synchronized to the basic information class, vehicle technical specification class, vehicle size and weight class, and vehicle static unstructured data in the static data type and the vehicle trajectory point information class, vehicle trajectory segment information class, vehicle usage information class, and vehicle dynamic unstructured data in the dynamic data type.

[0023] Preferably, in step S5, according to specific application scenarios and requirements, it is selected to create a static vehicle portrait using only static data, or to create a dynamic vehicle portrait using only dynamic data, or to create a total vehicle portrait using both static data and dynamic data.

[0024] Preferably, in step S1, if the collected data is a vehicle picture, the steps for constructing the vehicle portrait are as follows:

[0025] A1. Use convolutional neural network (CNN) to extract features of vehicle images. Input the provided vehicle images into the trained model to extract feature vectors.

[0026] A2. Assign corresponding weights according to the importance of attributes to vehicles, perform weighted processing on similarity measurement methods according to attribute weights, and search for the most similar vehicle information in the static database;

[0027] A3. If the threshold of the most similar vehicle record found is reasonable, search the dynamic database for a vehicle record that matches the vehicle record in the static database;

[0028] A4. Integrate static data information and dynamic data information to build a vehicle portrait.

[0029] Preferably, in step S1, if the collected data is a vehicle video, the steps for constructing the vehicle portrait are as follows:

[0030] B1. Extract vehicle features using convolutional neural networks (CNNs) or other methods, either by processing the video frame by frame and then aggregating the features of each frame, or by using a model specifically designed to process video data to extract the entire vehicle feature vector at once;

[0031] B2. Weighting the similarity measurement method according to the attribute weight, searching the static database and the dynamic database for the vehicle information most similar to the feature vector extracted from the video;

[0032] B3. Based on the comparison result of the similarity with the preset vehicle record threshold, select to add a new vehicle information record or update the static database and the dynamic database.

[0033] Preferably, in step S1, if the collected data is a vehicle picture, the steps for constructing the vehicle portrait are as follows:

[0034] C1. Extract business-related vehicle attribute data from static and dynamic databases and group them according to business requirements and data characteristics;

[0035] C2. Select an appropriate clustering algorithm, perform clustering analysis on the selected grouping, and divide the data points into different clusters;

[0036] C3. Based on the clustering results, create a vehicle profile for each cluster and display it using visualization tools.

[0037] The advantages and technical effects of the present invention are:

[0038] A vehicle portrait construction method based on operational data of the present invention obtains original vehicle data, determines the classification of the original vehicle data according to the attribute category of the original vehicle data, performs preprocessing operations on the original vehicle data of different classifications, and generates a database; obtains part of the label information through pictures or videos, performs knowledge calculation based on the constructed database to determine the missing labels of the corresponding vehicles, combines the determined missing labels of the vehicles and the extracted basic labels of the vehicles, and generates a static vehicle portrait or a dynamic vehicle portrait or a vehicle overall portrait according to business needs; groups the data according to business needs and data features, performs cluster analysis on the selected groups, divides the data points into different clusters, and creates a corresponding vehicle portrait for each cluster. By multi-classifying the original data, the efficiency of vehicle data management is improved, the security and confidentiality of the data are enhanced, more accurate data analysis and decision-making are supported, and data expansion and maintenance are facilitated. The existing labels extracted by the deep learning algorithm are combined with the labels of the existing database to generate static vehicle portraits, dynamic vehicle portraits, and overall vehicle portraits that meet business needs, or multiple clusters are obtained through clustering algorithms to generate vehicle portraits that meet business needs, thereby improving the efficiency of vehicle portrait creation. The created vehicle portraits provide data support for vehicle maintenance, optimization, and classification, and improve their accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flowchart of a vehicle portrait calculation method based on Internet of Vehicles operation data in the present invention;

[0040] Figure 2 A schematic diagram of the process of directly extracting task label information from a database to construct a vehicle portrait in the present invention;

[0041] Figure 3 A schematic diagram of the process of extracting vehicle information from a picture and combining it with a database to construct a vehicle portrait in the present invention;

[0042] Figure 4A schematic diagram of the process of extracting vehicle information from a video and combining it with a database to construct a vehicle portrait in the present invention;

[0043] Figure 5 The present invention is a flow chart of constructing vehicle portraits of different categories of vehicles by clustering method. DETAILED DESCRIPTION

[0044] In order to further understand the content, characteristics and effects of the present invention, the following embodiments are given for detailed description. It should be noted that the embodiments are illustrative, not restrictive, and the protection scope of the present invention cannot be limited thereby.

[0045] A vehicle portrait construction method based on operation data includes the following steps:

[0046] S1. Data collection: Collect various original operation data generated by the vehicle during operation;

[0047] S2. Data cleaning: delete unnecessary empty records in step S1, handle abnormal values, fill missing data, remove field duplication, handle error values, unify data formats, and standardize processing;

[0048] S3. Data integration: classify the cleaned label data in step S2 according to their structured degree, static data and dynamic data categories to form static structured data, static semi-structured data and static unstructured data; form dynamic structured data, dynamic semi-structured data and dynamic unstructured data, convert static semi-structured data into static structured data, convert dynamic semi-structured data into dynamic structured data, and integrate the cleaned and converted data;

[0049] S4. Create a database: add static data category information to the static database, add the above dynamic data category information to the dynamic database, add new tags to the database when receiving a tag building task, and update existing tags to the database when receiving a tag update task;

[0050] S5. According to the requirements for constructing vehicle portraits, construct a static vehicle portrait or a dynamic vehicle portrait or combine static data and dynamic data to construct a total vehicle portrait.

[0051] Preferably, the static data is further subdivided into: basic vehicle information, vehicle technical specifications, vehicle dimensions and weight, and vehicle static unstructured data; wherein:

[0052] Basic vehicle information: license plate, model, brand, color, and production date;

[0053] Vehicle technical specifications: fuel vehicle engine information, electric vehicle battery information, electric vehicle motor information, chassis and suspension, tires and brakes;

[0054] Vehicle size and weight: vehicle length, width, height, curb weight, and maximum load;

[0055] Vehicle static unstructured data: image data, video data, audio data, documents, and reports.

[0056] Preferably, the dynamic data is further subdivided into: vehicle trajectory point information, vehicle trajectory segment information, vehicle usage information and vehicle dynamic unstructured data;

[0057] Vehicle trajectory point information: location information, start and end time, start and end time of dangerous driving, total number of speeding alarms, number of sharp turns, number of sudden brakes, and number of sudden accelerations;

[0058] Track segment information: mileage, running time, speeding alarm mileage, maximum speeding alarm duration, fatigue driving mileage, maximum fatigue driving duration, number of fatigue driving times;

[0059] Vehicle usage information: fuel efficiency of fuel vehicles, battery performance of electric vehicles, parking location collection, vehicle fault and diagnostic information;

[0060] Vehicle dynamic unstructured data: log data, video data, audio data.

[0061] Preferably, in step S4, the operational data information that needs to be updated, deleted, or added is synchronized to the basic information class, vehicle technical specification class, vehicle size and weight class, and vehicle static unstructured data in the static data type and the vehicle trajectory point information class, vehicle trajectory segment information class, vehicle usage information class, and vehicle dynamic unstructured data in the dynamic data type.

[0062] Preferably, in step S5, according to specific application scenarios and requirements, it is selected to create a static vehicle portrait using only static data, or to create a dynamic vehicle portrait using only dynamic data, or to create a total vehicle portrait using both static data and dynamic data.

[0063] Preferably, in step S1, if the collected data is a vehicle picture, the steps for constructing the vehicle portrait are as follows:

[0064] A1. Use convolutional neural network (CNN) to extract features of vehicle images. Input the provided vehicle images into the trained model to extract feature vectors.

[0065] A2. Assign corresponding weights according to the importance of attributes to vehicles, perform weighted processing on similarity measurement methods according to attribute weights, and search for the most similar vehicle information in the static database;

[0066] A3. If the threshold of the most similar vehicle record found is reasonable, search the dynamic database for a vehicle record that matches the vehicle record in the static database;

[0067] A4. Integrate static data information and dynamic data information to build a vehicle portrait.

[0068] Preferably, in step S1, if the collected data is a vehicle video, the steps for constructing the vehicle portrait are as follows:

[0069] B1. Extract vehicle features using convolutional neural networks (CNNs) or other methods, either by processing the video frame by frame and then aggregating the features of each frame, or by using a model specifically designed to process video data to extract the entire vehicle feature vector at once;

[0070] B2. Weighting the similarity measurement method according to the attribute weight, searching the static database and the dynamic database for the vehicle information most similar to the feature vector extracted from the video;

[0071] B3. Based on the comparison result of the similarity with the preset vehicle record threshold, select to add a new vehicle information record or update the static database and the dynamic database.

[0072] Preferably, in step S1, if the collected data is a vehicle picture, the steps for constructing the vehicle portrait are as follows:

[0073] C1. Extract business-related vehicle attribute data from static and dynamic databases and group them according to business requirements and data characteristics;

[0074] C2. Select an appropriate clustering algorithm, perform clustering analysis on the selected grouping, and divide the data points into different clusters;

[0075] C3. Based on the clustering results, create a vehicle profile for each cluster and display it using visualization tools.

[0076] In order to more clearly illustrate the specific embodiments of the present invention, several embodiments are provided below:

[0077] Embodiment 1: Figure 2 As shown, the vehicle profiling method based on operation data of this embodiment can be implemented by an independent server or a server cluster composed of multiple servers, and the method includes:

[0078] S1. Data collection: Collect various original operation data generated by the vehicle during operation;

[0079] The collection methods include but are not limited to data information fed back in real time by vehicle-mounted sensors, information sent by mobile communication devices, and information uploaded by personnel. The raw data obtained through these channels usually covers multiple aspects of the vehicle. These raw data may include but are not limited to:

[0080] Basic vehicle information: license plate, model, brand, color, and production date;

[0081] Vehicle technical specifications: fuel vehicle engine information, electric vehicle battery information, electric vehicle motor information, chassis and suspension, tires and brakes;

[0082] Vehicle size and weight information: vehicle length, width, height, curb weight, and maximum load;

[0083] Vehicle trajectory point information: location information, start and end time, start and end time of dangerous driving, total number of speeding alarms, number of sharp turns, number of sudden brakes, and number of sudden accelerations;

[0084] Track segment information: mileage, running time, speeding alarm mileage, maximum speeding alarm duration, fatigue driving mileage, maximum fatigue driving duration, number of fatigue driving times;

[0085] Vehicle usage information: fuel efficiency for fuel vehicles, battery performance for electric vehicles, docking location collection, vehicle fault and diagnostic information.

[0086] In short, data collection is a crucial first step in the process of building a vehicle portrait based on operational data. Vehicle driving data can be obtained in real time by using on-board sensors, mobile communication devices, and personnel uploads. Multi-dimensional data such as vehicle driving status, road conditions, driver behavior, and vehicle performance parameters are captured in real time and comprehensively. In order to ensure that these real-time collected data can be effectively used, the collected data is transmitted to the server through mobile communication networks (such as 4G, 5G and other wireless communication technologies) or near-field communication technologies (such as WIFI, Bluetooth, etc.).

[0087] S2. Data cleaning: Ensure the accuracy, completeness and consistency of data, including:

[0088] Delete unnecessary empty records: In the data set, there may be records that are completely blank or missing key information. These records have no value for subsequent analysis and modeling, and may even introduce noise. Therefore, it is necessary to identify and delete these unnecessary empty records;

[0089] Handling outliers: Outliers are values ​​in a data set that are significantly different from other data points. They may be caused by errors, measurement errors, or data entry errors. Different outlier handling methods can be selected according to different data, including direct deletion, replacement (replacement with mean, median, mode, etc.), grouping (grouping data and handling outliers separately) or using statistical methods (such as distribution-based methods, Z-score method, etc.) to judge and handle outliers;

[0090] Filling missing data: There may be missing values ​​in the data set, which may be caused by various reasons, such as data loss, incomplete records, etc. Different methods of filling missing values ​​can be selected according to different data, including mean filling (for numerical features), mode filling (for categorical features), interpolation methods (such as linear interpolation, polynomial interpolation, etc.) and prediction model filling (such as using KNN, decision tree, random forest and other models to predict missing values);

[0091] Field deduplication: If there are duplicate records or fields in the data set, deduplication is required;

[0092] Handling error values: Identify and correct error values. The method for handling error values ​​depends on the specific type and cause of the error. It can be corrected by simple replacement (such as replacing the error value with a reasonable default value); or it can be detected and corrected by complex algorithms or models;

[0093] Unified data format: To facilitate subsequent analysis and processing, the data format needs to be unified into a consistent standard format;

[0094] Data standardization: Eliminate the dimensional differences between different features. Standardization is usually achieved using the Z-score method or the Min-Max method.

[0095] S3. Data integration: Classify tags according to their degree of structure and static and dynamic categories to form static structured data, static semi-structured data, and static unstructured data; form dynamic structured data, dynamic semi-structured data, and dynamic unstructured data, convert static semi-structured data into static structured data, convert dynamic semi-structured data into dynamic structured data, and integrate the cleaned and converted data.

[0096] S4. Create a static database and a dynamic database: add static data category information to the static database, and add dynamic data category information to the dynamic database.

[0097] S5. Create vehicle portrait: Obtain label information from the static database and the dynamic database. According to business needs, only static vehicle portraits are constructed through static data, or only dynamic vehicle portraits are constructed through dynamic data, or a total vehicle portrait is constructed through static and dynamic data and a visual report is formed.

[0098] Embodiment 2: Figure 3 As shown, if a static database and a dynamic database have been created, there is no need to create them again. Otherwise, it is necessary to perform the steps of data collection, data cleaning, data integration, and creation of a static database and a dynamic database in the first embodiment.

[0099] If the current business provides vehicle images and requires the vehicle portrait of the business, the following steps need to be performed:

[0100] S1. Extract vehicle information from the image

[0101] Preprocessing: resizing, standardization, normalization, denoising, etc.

[0102] The standardization formula is as follows:

[0103]

[0104] Where μ is the mean of the image, x represents the image matrix, σ represents the standard deviation, and N represents the number of pixels in the image.

[0105] The normalization formula is as follows:

[0106]

[0107] where x i Represents the image pixel value, min(x) represents the minimum value of the image pixel, and max(x) represents the maximum value of the image pixel.

[0108] Reasoning: Use convolutional neural network (CNN) to extract features of vehicle images, input the provided vehicle images into the trained model, extract feature vectors, and obtain basic label information of the vehicle. Use YOLO-based multi-dimensional feature recognition to extract feature vectors such as license plate, vehicle color, vehicle brand, vehicle model, and vehicle logo.

[0109] S2. Find the most similar vehicle information in the database

[0110] According to the importance of vehicle attributes, each attribute is assigned a corresponding weight w = [w1, w2, w3, ..., w n], the feature vector extracted from the image is weighted according to the weight of the attribute, and the similarity measurement method is weighted. The weighted similarity measurement method is used to find the vehicle information most similar to the input vehicle image in the static database, and the vehicle record with the highest similarity is found as the most similar vehicle.

[0111]

[0112] where w i Represents the weight of the i-th attribute, satisfying

[0113] where s i It is the similarity of the vehicle in the i-th attribute. Its calculation method depends on the specific type and measurement method of the attribute. When quantitative attributes such as license plate number are extracted, the matching vehicle information can be directly obtained from the static database.

[0114] S3. Similarity comparison

[0115] The calculated similarity is compared with the preset threshold. If the similarity does not exceed the threshold, a new vehicle information record is created in the static database; if the threshold is reasonable, the static database is updated and the dynamic database is searched for vehicle information that matches the vehicle record in the static database.

[0116] S4. Create vehicle profile

[0117] The label information of the most similar vehicle found in the static database, the matching vehicle label information found in the dynamic database, and the vehicle label information extracted from the image are integrated to finally form a complete vehicle portrait and a visual report. If the business requirement is to create only a static vehicle portrait, there is no need to search the dynamic database for vehicle information that matches the vehicle record in the static database. The label information of the most similar vehicle found in the static database and the vehicle label information extracted from the image are directly integrated to form a static vehicle portrait and a visual report.

[0118] Embodiment 3: Figure 4 As shown, if a static database and a dynamic database have been created, there is no need to create them again. Otherwise, you need to perform the steps of data collection, data cleaning, data integration, and creation of static databases and dynamic databases in Example 1. If the current business provides a vehicle video and requires obtaining a vehicle portrait of the vehicle, you need to perform the following steps:

[0119] S1. Vehicle information extraction in video

[0120] The features of a vehicle can be extracted using a convolutional neural network (CNN), by processing the video frame by frame and then aggregating the features of each frame, or by using a model specially designed to process video data to extract the entire vehicle feature vector at one time to obtain the basic label information of the vehicle.

[0121] S2. Find the most similar vehicle information in the database

[0122] According to the importance of vehicle attributes, a corresponding weight is assigned to each attribute. For the feature vector extracted from the video, the similarity measurement method is weighted according to the weight of the attribute. The weighted similarity measurement method is used to find the vehicle label information most similar to the feature vector extracted from the video in the static database and the dynamic database, and the vehicle record with the highest similarity is found as the most similar vehicle.

[0123] S3. Similarity comparison

[0124] The calculated similarity is compared with a preset threshold. If the similarity does not exceed the threshold, a new vehicle information record is created in the static database and the dynamic database. This record should contain the feature vector extracted from the video and any other attribute information that can be identified or obtained from the video or user input. If the similarity exceeds the threshold and the attribute information of the database is incomplete or needs to be updated, the attribute information extracted from the video is updated to the static database and the dynamic database.

[0125] Integrate the vehicle label information found in the static database, the vehicle label information found in the dynamic database, and the vehicle label information extracted from the video to form a complete vehicle portrait and a visual report. If the business requires only the creation of a static vehicle portrait, then only the label information of the most similar vehicle found in the static database and the static vehicle label information extracted from the video need to be integrated to form a static vehicle portrait and a visual report. If the business requires only the creation of a dynamic vehicle portrait, then only the label information of the most similar vehicle found in the dynamic database and the dynamic vehicle label information extracted from the video need to be integrated to form a dynamic vehicle portrait and a visual report.

[0126] Embodiment 4: Figure 5 As shown, if the static database and the dynamic database have been created before, there is no need to create them again. Otherwise, it is necessary to perform the steps of data collection, data cleaning, data integration, and creation of static database and dynamic database in Example 1.

[0127] If the current business requires drawing different vehicle portraits according to different categories, the following steps need to be performed:

[0128] S1. Data preparation:

[0129] Extract vehicle-related attribute data from static and dynamic databases and perform preprocessing operations on the data, including:

[0130] Data cleaning: remove duplicate data, missing values, outliers or noisy data;

[0131] Data integration: Integrate relevant data from static and dynamic databases into one data set;

[0132] Feature engineering: Extract or create new features based on business needs.

[0133] S2. Data Grouping

[0134] Determine grouping criteria: Based on business requirements and data characteristics, determine one or more grouping criteria and group the data based on the selected characteristics.

[0135] S3. Clustering

[0136] Select an appropriate clustering algorithm based on the characteristics of the data and business requirements. For vehicle data, the clustering algorithms that can be selected include K-means, DBSCAN, hierarchical clustering, etc. Perform clustering analysis on the selected groups to divide the data into different clusters. Input the preprocessed data into the clustering algorithm and execute the clustering process. The algorithm will divide the data points into different clusters based on the similarity of the data and the set parameters. Based on the clustering results, define a vehicle profile for each cluster and generate a visual report to display the clustering results and vehicle profile.

[0137] K-means clustering algorithm steps:

[0138] Initialization: Randomly select K data points as the initial centroids (category centers).

[0139] Assign data points: For each data point in the dataset, calculate its distance from the K centroids and assign it to the category where the nearest centroid is located. The distance calculation formula is as follows:

[0140]

[0141] Where x is the data point, c i is the i-th cluster center, d is the dimension of the data, x i and c ij are x and c respectively i The value in the jth dimension.

[0142] Update Centroids: Recalculate the centroid of each class.

[0143] Iteration: Repeat the steps of assigning data points and updating centroids until the stopping condition is met (the centroids no longer change or the maximum number of iterations is reached).

[0144] In summary, the present invention selects corresponding modules to construct vehicle portraits according to business requirements, obtains original vehicle data, determines the classification of original vehicle data according to the attribute categories of original vehicle data, performs preprocessing operations on original vehicle data of different classifications, and generates a database; obtains partial label information through pictures or videos, performs knowledge calculation based on the constructed database to determine the missing labels of the corresponding vehicles, combines the determined missing labels of the vehicles and the extracted basic labels of the vehicles, and generates static vehicle portraits or dynamic vehicle portraits or overall vehicle portraits that meet business requirements; groups data according to business requirements and data features, performs cluster analysis on selected groups, divides data points into different clusters, and creates corresponding vehicle portraits for each cluster. By multi-classifying the original data, the efficiency of vehicle data management is improved, the security and confidentiality of the data are enhanced, more accurate data analysis and decision-making are supported, and data expansion and maintenance are facilitated. The existing labels extracted by the deep learning algorithm are combined with the labels of the existing database to generate static vehicle portraits, dynamic vehicle portraits, and overall vehicle portraits that meet business needs, or multiple clusters are obtained through clustering algorithms to generate vehicle portraits that meet business needs, thereby improving the efficiency of vehicle portrait creation. The created vehicle portraits provide data support for vehicle maintenance, optimization, and classification, and improve their accuracy.

[0145] Finally, all the parts not described in the present invention adopt mature products and mature technical means in the prior art.

[0146] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should fall within the scope of protection of the appended claims of the present invention.

Claims

1. A vehicle portrait construction method based on operation data, It is characterized in that The following steps are involved: S1. Data collection: Collect various original operation data generated by the vehicle during operation; S2. Data cleaning: delete unnecessary empty records in step S1, handle abnormal values, fill missing data, remove field duplication, handle error values, unify data format, and standardize processing; S3. Data integration: Classify the label data cleaned in step S2 according to its structured degree, static data and dynamic data categories to form static structured data, static semi-structured data and static unstructured data; Form dynamic structured data, dynamic semi-structured data, and dynamic unstructured data, convert static semi-structured data into static structured data, convert dynamic semi-structured data into dynamic structured data, and integrate the cleaned and converted data; S4. Create a database: add static data category information to the static database, add the above dynamic data category information to the dynamic database, add new tags to the database when receiving a tag building task, and update existing tags to the database when receiving a tag update task; S5. According to the requirements for constructing vehicle portraits, construct a static vehicle portrait or a dynamic vehicle portrait or combine static data and dynamic data to construct a total vehicle portrait.

2. A vehicle portrait construction method based on operation data according to claim 1, Features: The static data is further subdivided into: basic vehicle information, vehicle technical specifications, vehicle dimensions and weight, and vehicle static unstructured data; among which: Basic vehicle information: license plate, model, brand, color, and production date; Vehicle technical specifications: fuel vehicle engine information, electric vehicle battery information, electric vehicle motor information, chassis and suspension, tires and brakes; Vehicle size and weight: vehicle length, width, height, curb weight, and maximum load; Vehicle static unstructured data: image data, video data, audio data, documents, and reports.

3. A vehicle portrait construction method based on operation data according to claim 1, Features: The dynamic data is further subdivided into: vehicle trajectory point information, vehicle trajectory segment information, vehicle usage information and vehicle dynamic unstructured data; Vehicle trajectory point information: location information, start and end time, start and end time of dangerous driving, total number of speeding alarms, number of sharp turns, number of sudden brakes, and number of sudden accelerations; Track segment information: mileage, running time, speeding alarm mileage, maximum speeding alarm duration, fatigue driving mileage, maximum fatigue driving duration, number of fatigue driving times; Vehicle usage information: fuel efficiency of fuel vehicles, battery performance of electric vehicles, parking location collection, vehicle fault and diagnostic information; Vehicle dynamic unstructured data: log data, video data, audio data.

4. A vehicle portrait construction method based on operation data according to claim 1, Features: In step S4, the operational data information that needs to be updated, deleted, or added is synchronized to the basic information class, vehicle technical specification class, vehicle size and weight class, and vehicle static unstructured data in the static data type and the vehicle trajectory point information class, vehicle trajectory segment information class, vehicle usage information class, and vehicle dynamic unstructured data in the dynamic data type.

5. A vehicle portrait construction method based on operation data according to claim 1, Features: In step S5, according to specific application scenarios and requirements, it is selected to create a static vehicle portrait using only static data, or to create a dynamic vehicle portrait using only dynamic data, or to create a total vehicle portrait using both static data and dynamic data.

6. A vehicle portrait construction method based on operation data according to claim 1, Features: In step S1, if the collected data is a vehicle picture, the steps for constructing a vehicle portrait are as follows: A1. Use convolutional neural network (CNN) to extract features of vehicle images. Input the provided vehicle images into the trained model to extract feature vectors. A2. Assign corresponding weights according to the importance of attributes to vehicles, perform weighted processing on similarity measurement methods according to attribute weights, and search for the most similar vehicle information in the static database; A3. If the threshold of the most similar vehicle record found is reasonable, search the dynamic database for a vehicle record that matches the vehicle record in the static database; A4. Integrate static data information and dynamic data information to build a vehicle portrait.

7. A vehicle portrait construction method based on operation data according to claim 1, Features: In step S1, if the collected data is a vehicle video, the steps for constructing a vehicle portrait are as follows: B1. Extract vehicle features using convolutional neural networks (CNNs) or other methods, either by processing the video frame by frame and then aggregating the features of each frame, or by using a model specifically designed to process video data to extract the entire vehicle feature vector at once; B2. Weighting the similarity measurement method according to the attribute weight, searching the static database and the dynamic database for the vehicle information most similar to the feature vector extracted from the video; B3. Based on the comparison result of the similarity with the preset vehicle record threshold, select to add a new vehicle information record or update the static database and the dynamic database.

8. A vehicle portrait construction method based on operation data according to claim 1, Features: In step S1, if the collected data is a vehicle picture, the steps for constructing a vehicle portrait are as follows: C1. Extract business-related vehicle attribute data from static and dynamic databases and group them according to business requirements and data characteristics; C2. Select an appropriate clustering algorithm, perform clustering analysis on the selected grouping, and divide the data points into different clusters; C3. Based on the clustering results, create a vehicle profile for each cluster and display it using visualization tools.

Images