Method for collecting and processing automobile data

Abstract

The invention relates to the technical field of internet-of-things automobiles, in particular to an automobile data collecting and processing method which comprises the steps that an internet-of-things module is loaded behind a vehicle standard OBD diagnosis port, a vehicle bus data flow is collected and uploaded to a cloud background, and according to all scene requirements in the whole life cycle of a vehicle, the internet-of-things module is connected with the internet-of-things module; and calling an algorithm model to analyze, analyze and process the Internet of Vehicles data flow, actively developing user demands, analyzing user intentions to adjust data weights, and completing data interaction with a user side. According to the automobile data collecting and processing method, the control logic and the AI algorithm are fused, the threshold value judgment rule of a power assembly and a new energy electric drive assembly system is improved, the strong nonlinear fitting capability of XGBoost is combined, the real-time performance and reliability of automobile condition monitoring are improved, XGBoost-Boruta mixed feature screening is adopted, and the real-time performance and reliability of automobile condition monitoring are improved. Compared with a traditional Dropout method, the model interpretation and generalization ability are improved, a feature selection optimization strategy is achieved, and dynamic updating of a data acquisition model of a departure end under a background and millisecond-level response of background calculation are supported through incremental training and a low-delay architecture and based on a distributed data acquisition and calculation framework of the high-speed Internet of Things.

Description

Technical Field

[0001] The present invention relates to the technical field of Internet of Things vehicles, and specifically to a method for collecting and processing vehicle data. Background Art

[0002] During the operation of current motor vehicles, construction machinery, and ships, data such as the overall machine operation data, exhaust emission monitoring data, and power battery performance data are continuously generated. The existing stock of more than 400 million devices in China is an existing Internet of Things terminal that can provide high-quality data. Taking automobiles, which have the largest stock, as an example, vehicles continuously generate a huge amount of data at high frequency during operation, including aspects such as the vehicle itself, driver operation, road environment, and autonomous driving. By fully collecting high-value data from devices such as vehicles and combining it with application scenarios for data analysis and value mining, it not only meets the requirements of the development of the digital economy era, but also the collected data can be used to facilitate the lives of vehicle owners, accumulate vehicle owner data assets, and can be used for technology research and development and trust transparency in the vehicle value chain. Therefore, a data collection and processing method for motor vehicles, construction machinery, and ships of various brands needs to be designed.

[0003] After retrieval, a new energy vehicle data collection and processing method and system is disclosed in the invention patent with Chinese patent number CN113858956A. The vehicle data collection and processing method in this invention patent is based on the historical messages of in-vehicle data collection terminals. During the operation of the vehicle, by comprehensively using the real-time and historical operation data of the vehicle, through analyzing the coupling characteristics of power battery characteristic parameters and potential safety hazards, the safe operation of the power battery is effectively monitored. With the help of characteristic parameters, it is beneficial to timely discover and evaluate the problems of potential safety hazards caused by power battery mutations, improving the safety and reliability of power battery use. The results of the safety assessment of the vehicle are uploaded to the platform, and then the operation situation of the vehicle can be evaluated. The vehicle can be processed or detected according to the algorithm results;

[0004] However, the new energy vehicle data collection and processing method and system only focuses on the processing of single data dimension related to power batteries, relying on static rules resulting in a fixed parameter system that cannot be iteratively updated dynamically through the cloud, lacking a dynamic learning mechanism and being unable to perform real-time diagnosis. Therefore, a method for collecting and processing vehicle data is proposed. Summary of the Invention

[0005] (1) Technical Problems to be Solved

[0006] In view of the deficiencies of the prior art, the present invention provides a method for collecting and processing vehicle data, which has the advantage of realizing iterative dynamic update through the cloud based on a dynamic learning mechanism, and solves the problems that the existing vehicle data collection and processing method solidifies the data parameter system and cannot achieve real-time diagnosis and dynamic update in the above background art.

[0007] (2) Technical Solution

[0008] To achieve the above-mentioned combined subjective and objective double assignment method, which can reduce the influence of subjective randomness and reflect the information volume characteristics provided by the data itself, for the purpose of reducing the weight fluctuation caused by data changes, the present invention provides the following technical solution: A method for collecting and processing vehicle data, including the following implementation steps:

[0009] S1. Data collection: By installing a vehicle networking module behind the vehicle standard OBD diagnostic port, collecting the vehicle bus data stream and uploading it to the cloud background;

[0010] S2. Data analysis: The cloud background analyzes and processes the vehicle networking data stream based on an algorithm model, and deeply analyzes the data in combination with the new content of the industry reference database and the industry knowledge base;

[0011] S3. Weight adjustment: Develop potential requirements based on user behavior, actively analyze user intentions and adjust data weights;

[0012] S4. Data interaction feedback: The cloud background transmits the analysis result back to the customer-specified interface or the user terminal interface according to the user request, and realizes data interaction and operation guidance based on the AI model.

[0013] Preferably, the vehicle networking module is adapted to the vehicle standard OBD diagnostic port, and can be equipped with an independent Internet of Things module to directly connect to the background, or can be connected to the cloud background data after being connected to the in-vehicle streaming media mirror or the user's mobile phone through Bluetooth communication. The specific steps of data collection include:

[0014] 1) Connect the vehicle networking module to the vehicle-end OBD diagnostic port, and the cloud background dynamically issues data interaction instructions according to the vehicle model and scenario function requirements;

[0015] 2) The OBD adapter in the vehicle networking module converts the data interaction instruction into a vehicle interaction instruction to obtain the vehicle bus information, and the OBD adapter simultaneously converts the vehicle bus information into a vehicle networking data stream;

[0016] 3) The vehicle networking device uploads the vehicle networking data stream to the cloud background through the Internet of Things, Bluetooth, and other wired connections;

[0017] The cloud background for data collection and processing accesses the user client through the APP software module or the API interface, and is used to output the obtained vehicle networking data stream and the data diagnosis report to the user, and defines the data collection range and collection frequency according to the actual usage scenario requirements and vehicle model support situation. The highest data collection frequency is 20HZ.

[0018] Preferably, the steps for the cloud background of vehicle data collection and processing to analyze and process the vehicle networking data stream include:

[0019] 1) The cloud backend parses and self-learns the obtained Internet of Vehicles data stream.

[0020] The data collected in the cloud database all carry hexadecimal numbers with timestamps. The cloud backend automatically parses and processes them according to requirements, including: a. Parsing environmental protection data related to vehicle emissions; b. Parsing data related to vehicle failures; c. Parsing data on the performance of the whole vehicle and key components in combination with vehicle brands and model characteristics.

[0021] 2) Based on the forward diagnosis algorithm and the correlation algorithm of XGBoost, define monitoring fields and thresholds according to scenario requirements, and call relevant calculation models when preset conditions are met.

[0022] 3) AI integrates and forms a report and outputs it by synthesizing the results of multiple different calculation models.

[0023] The steps for in-depth analysis of data by combining the industry reference database and the new content of the industry knowledge base include:

[0024] 1) Connect to the government open platform, third-party databases, and industry research reports to access the industry reference database, write into the knowledge base rule engine according to the new rules, and update the domain ontology model, and add entity relationships in the knowledge graph.

[0025] 2) Map the industry database and the knowledge base to achieve heterogeneous data fusion.

[0026] 3) Extract domain features from CAN bus data in combination with the knowledge base, use the Apriori algorithm to mine association rules, perform rule constraint learning based on physical models and data-driven, and build a knowledge-enhanced hybrid model.

[0027] 4) Establish an online learning mechanism to detect network attack patterns in real time and achieve incremental knowledge update.

[0028] Preferably, analyze data based on the forward diagnosis algorithm of the data stream. The specific steps include:

[0029] 1) Obtain the timing data of the whole vehicle and the engine, or the control logic of the electric drive system in real time, including parameter data such as engine speed, temperature, or drive motor speed, power battery voltage and temperature, etc., and collect the data stream in real time based on the low-latency transmission of high-speed Internet of Things communication technology.

[0030] 2) Use the state machine model to preset the control logic threshold to achieve control logic anomaly detection, including:

[0031] a. Based on engine operating conditions judgment: if (speed > threshold and oil pressure < threshold),

[0032] then trigger a diagnostic signal;

[0033] b. Verify the robustness of the logic rules based on the hardware device in the HIL test platform;

[0034] 3) Output the fault level based on the diagnostic rule base, combine with the finite state machine FSM to implement multi-condition joint judgment, and generate a fault and special situation warning mechanism for real-time diagnostic decision-making, including:

[0035] a. Determine all possible fault states and input conditions to be processed to clarify the diagnostic requirements and conditions, identify and list all condition combinations that need to be jointly judged, and define the state behavior by enumerating all states;

[0036] b. Build the transition relationship by defining possible transition paths and corresponding condition combinations for each state, describe the state transition logic using a transition table or UML state diagram, use logical operators to combine conditions to implement multi-condition processing, and if multiple transition conditions are satisfied simultaneously, define the priority or adjust the conditions;

[0037] c. Represent the current state variable using enumeration or constants, trigger state checks when conditions change through event-driven, detect conditions regularly according to the polling mechanism, and build the finite state machine FSM structure.

[0038] Preferably, analyze the data based on the correlation algorithm of XGBoost, and the specific steps include:

[0039] 1) Input the vehicle accident data as the first period and the vehicle operation data as the second period based on the obtained vehicle networking data stream, and perform data preprocessing operations, including missing value filling, data standardization, and feature alignment;

[0040] 2) Adopt the XGBoost-Boruta hybrid strategy, conduct a preliminary screening of feature importance based on the Boruta algorithm, optimize feature selection and eliminate redundant variables, and build an XGBoost model, where the gain calculation formula of XGBoost is expressed as:

[0041]

[0042] where G and H are the first / second derivatives of the loss function, and λ and γ are regularization parameters;

[0043] 3) Build the objective function, expressed as:

[0044]

[0045] where is used to control the complexity of the tree, train the XGBoost model based on the incremental training strategy, retain the prediction results of the previous t - 1 rounds in each round, and add a new function (f t ) to optimize the residual;

[0046] 4) Verify the model accuracy based on the 1DCNN comparison method, and test the robustness on the HIL platform in combination with real-time vehicle data. The accuracy rate can reach 95% in the reclassification scenario, and multi-label output of fault types is supported.

[0047] Preferably, the specific steps for analyzing user needs and intentions and realizing data weight adjustment are as follows:

[0048] 1) Based on LP keyword extraction and session tree analysis and processing, analyze the explicit active demand data of users, and adjust the data weight range to 0.8 - 1.0; based on LSTM behavior sequence modeling and processing, analyze the implicit potential demand of users, and adjust the data weight range to 0.5 - 0.7; based on collaborative filtering similarity matching and processing, analyze the passive trigger demand of users, and adjust the data weight range to 0.3 - 0.5;

[0049] 2) Parse the text intention according to the BERT model, calculate the page area attention through eye movement tracking data, and construct a Markov model of the user operation chain to identify and analyze the user behavior time sequence;

[0050] 3) Dynamically adjust the weight according to the weight calculation framework, expressed as:

[0051] W i = α·I explicit + β·I implicit + γ·R context

[0052] where α is the explicit intention factor, β is the implicit intention factor, and γ is the environmental factor.

[0053] Preferably, the steps for realizing data interaction include:

[0054] 1) Construct a modular system architecture and select the core components of the ASR module. Strengthen the voice data through noise injection and accent enhancement, and optimize the recognition accuracy based on context-aware ASR through domain adaptation technology, expressed as: where is the domain loss term, use GST to realize the dynamic adjustment of intonation, and embed the AI voice model in the user interface;

[0055] 2) The user interface includes the mobile APP, in-vehicle all-in-one machine, and streaming media rearview mirror. Access the diagnostic report based on the APP module and the program web page, and the AI voice model explains in real time according to the diagnostic report.

[0056] (III) Beneficial effects

[0057] Compared with the prior art, the present invention provides a method for collecting and processing automotive data, having the following beneficial effects:

[0058] The method for collecting and processing vehicle data combines automotive technical logic with AI algorithms based on big data in the automotive industry, combines traditional vehicle condition threshold judgment rules with the strong non-linear fitting ability of XGBoost, improves the real-time performance and reliability of diagnosis, adopts XGBoost-Boruta hybrid feature screening, improves the model interpretability and generalization ability compared with the traditional Dropout method, realizes the feature selection optimization strategy, and through incremental training and low-latency architecture, a distributed data acquisition and computing framework based on high-speed Internet of Things supports the dynamic update of the vehicle-end data acquisition model sent by the background and the millisecond-level response of background computing. Brief Description of the Drawings

[0059] Figure 1 It is a flowchart of the method for the cloud background of the present invention to collect and process data of Internet of Things devices;

[0060] Figure 2 It is a flowchart of the AI Agent data processing workflow of the present invention;

[0061] Figure 3 It is a flowchart of the AI autonomous in-depth analysis workflow of the present invention;

[0062] Figure 4 It is Figure 3 a schematic diagram of the data input block in

[0063] Figure 5 It is Figure 3 a schematic diagram of the special behavior block of the driver scene in

[0064] Figure 6 It is Figure 3 a schematic diagram of the vehicle performance, traffic environment and vehicle fault scenario blocks in

[0065] Figure 7 It is Figure 3 a schematic diagram of the AI model self-decision block in

[0066] Figure 8 It is a schematic diagram of the data acquisition link based on the in-vehicle network device of the present invention;

[0067] Figure 9 It is a schematic diagram of the data acquisition link based on the streaming media rearview mirror network device of the present invention. Detailed Embodiments

[0068] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the embodiments and the accompanying drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0069] Embodiment 1:

[0070] S1. Data acquisition: By installing a vehicle networking module behind the standard OBD diagnostic port of the vehicle, the vehicle bus data stream is collected and uploaded to the cloud background.

[0071] Furthermore, the vehicle networking module is adapted to the standard OBD diagnostic port of the vehicle. It can have an independent Internet of Things module to directly connect to the background, or it can be connected to the cloud background data after being connected to the in-vehicle streaming mirror or the user's mobile phone through Bluetooth communication. The specific steps of data acquisition include:

[0072] 1) Connect the vehicle networking module to the vehicle-end OBD diagnostic port, and the cloud background dynamically issues data interaction instructions according to the vehicle model and scenario function requirements;

[0073] 2) The OBD adapter in the vehicle networking module converts the data interaction instructions into vehicle interaction instructions to obtain vehicle bus information, and at the same time converts the vehicle bus information into a vehicle networking data stream;

[0074] 3) The vehicle networking device uploads the vehicle networking data stream to the cloud background through the Internet of Things, Bluetooth, and other wired connections;

[0075] Data aggregation and processing: The cloud background accesses the client through the APP software module or the API interface, and is used to output the obtained vehicle networking data stream and the data diagnosis report to the client. The data acquisition range and acquisition frequency are defined according to the actual usage scenario requirements and vehicle model requirements, and the highest data acquisition frequency is 20HZ.

[0076] Specifically, connect the after-installed acquisition device to the OBD Bluetooth interface. The extreme vehicle networking module is connected to the OBD Bluetooth interface and the cloud background data through Bluetooth, collects vehicle bus information and converts it into a vehicle networking data stream, and analyzes the vehicle networking data.

[0077] Embodiment 2:

[0078] S2. Data analysis: The cloud background analyzes and processes the vehicle networking data stream based on the algorithm model, and deeply analyzes the data in combination with the new content of the industry reference database and the industry knowledge base.

[0079] Furthermore, the steps for the cloud background of vehicle data aggregation and processing to analyze and process the vehicle networking data stream based on the algorithm model include:

[0080] 1) The cloud backend parses and self-learns the obtained vehicle networking data stream.

[0081] The data collected in the cloud database all carry hexadecimal numbers with timestamps. The cloud backend automatically parses and processes them according to requirements, including: a. Parsing environmental protection data related to vehicle emissions; b. Parsing data related to vehicle failures; c. Parsing data on the performance of the whole vehicle and key components in combination with vehicle brands and model characteristics.

[0082] 2) Based on the forward diagnostic algorithm and the correlation algorithm of XGBoost, define monitoring fields and thresholds according to scenario requirements, and call relevant calculation models when preset conditions are met.

[0083] 3) Synthesize the results of multiple different calculation models, and the AI integrates and forms a report and outputs it.

[0084] The steps for in-depth analysis of data by combining the industry reference database and the new content of the industry knowledge base include:

[0085] 1) Connect to the government open platform, third-party databases, and industry research reports to access the industry reference database, write rules into the knowledge base rule engine according to new rules, and update the domain ontology model, and add entity relationships in the knowledge graph.

[0086] 2) Map the industry database and the knowledge base to achieve heterogeneous data fusion.

[0087] 3) Extract domain features from CAN bus data in combination with the knowledge base, use the Apriori algorithm to mine association rules, perform rule constraint learning based on physical models and data-driven, and build a knowledge-enhanced hybrid model.

[0088] 4) Establish an online learning mechanism to detect network attack patterns in real time and achieve incremental knowledge update.

[0089] Analyze data based on the forward diagnostic algorithm of the data stream. The specific steps include:

[0090] 1) Obtain the timing data of the whole vehicle and the engine, or the control logic of the electric drive system in real time, including parameter data such as engine speed, temperature, or drive motor speed, power battery voltage and temperature, etc., and collect the data stream in real time based on the low-latency transmission of high-speed Internet of Things communication technology.

[0091] 2) Use the state machine model to preset the control logic threshold to achieve control logic anomaly detection, including:

[0092] a. Based on engine condition judgment: if (speed > threshold and oil pressure < threshold),

[0093] then trigger a diagnostic signal;

[0094] b. Verify the robustness of the logic rules on the HIL test platform based on the hardware device;

[0095] 3) Output the fault level based on the diagnostic rule base, combine with the finite state machine FSM to implement multi-condition joint judgment, and generate a fault and special situation warning mechanism for real-time diagnostic decision-making, including:

[0096] a. Determine all possible fault states and input conditions to be processed to clarify the diagnostic requirements and conditions, identify and list all condition combinations that need to be jointly judged, and define the state behavior by enumerating all states;

[0097] b. Build the transition relationship by defining possible transition paths and corresponding condition combinations for each state, describe the state transition logic using a transition table or UML state diagram, use logical operators to combine conditions to implement multi-condition processing, and if multiple transition conditions are satisfied simultaneously, define the priority or adjust the conditions;

[0098] c. Represent the current state variable using enumeration or constants, trigger state checks when conditions change through event-driven, regularly detect conditions according to the polling mechanism, and build the finite state machine FSM structure;

[0099] Analyze data using the correlation algorithm based on XGBoost. The specific steps include:

[0100] 1) Input vehicle accident data as the first period and vehicle operation data as the second period based on the obtained vehicle networking data stream, and perform data preprocessing operations, including missing value filling, data standardization, and feature alignment;

[0101] 2) Adopt the XGBoost-Boruta hybrid strategy, conduct a preliminary screening of feature importance based on the Boruta algorithm, optimize feature selection and eliminate redundant variables, and build an XGBoost model. The gain calculation formula of XGBoost is expressed as:

[0102]

[0103] where G and H are the first / second derivatives of the loss function, and λ and γ are regularization parameters;

[0104] 3) Build the objective function, expressed as:

[0105]

[0106] where is used to control the complexity of the tree, train the XGBoost model based on the incremental training strategy, retain the prediction results of the previous t - 1 rounds in each round, and add a new function (f t ) to optimize the residual;

[0107] 4) Verify the model accuracy based on the 1DCNN comparison method, and test the robustness on the HIL platform by combining real-time vehicle data. The accuracy rate can reach 95% in the reclassification scenario, and it supports multi-label output of fault types.

[0108] Specifically, combining the forward diagnosis algorithm with the correlation algorithm of XGBoost, calculate, analyze and learn the vehicle networking data stream. The forward diagnosis algorithm can combine the data stream with the vehicle and engine control logic, and obtain the timing data of the vehicle and engine control logic in real time through in-vehicle sensors and the electronic control unit ECU, detect abnormalities in the control logic, and achieve multi-condition joint judgment and real-time diagnosis decision according to the fault level;

[0109] Adopting the correlation algorithm of XGBoos of big data and AI, it can combine the consistency of vehicle quality and horizontally compare case data, optimize the selected features by using the XGBoost-Boruta hybrid strategy, analyze the output results according to the XGBoost model and verify through 1DCNN comparison.

[0110] Embodiment 3:

[0111] S3. Weight adjustment: Develop potential requirements based on user behavior, actively analyze user intentions and adjust data weights.

[0112] Furthermore, the specific steps to analyze user needs and intentions and implement data weight adjustment are as follows:

[0113] 1) Based on LP keyword extraction and conversation tree analysis and processing, analyze the user's explicit active demand data, and adjust the data weight range to 0.8 - 1.0; based on LSTM behavior sequence modeling and processing, analyze the user's implicit potential needs, and adjust the data weight range to 0.5 - 0.7; based on collaborative filtering similarity matching processing, analyze the user's passive triggered needs, and adjust the data weight range to 0.3 - 0.5;

[0114] 2) Parse the text intention according to the BERT model, calculate the page area attention through eye movement tracking data, and construct a Markov model of the user operation chain to identify and analyze the user behavior timing;

[0115] 3) Dynamically adjust the weight according to the weight calculation framework, expressed as:

[0116] W i = α·I explicit + β·I implicit + γ·R context

[0117] where α is the explicit intention factor, β is the implicit intention factor, and γ is the environmental factor.

[0118] Specifically, by quantifying and differentiating the demand intensity through weight intervals, designing a dynamic range of resource allocation where high-value explicit demands are greater than passive demands can avoid weight solidification. Based on an elastic weight calculation framework, dynamically optimizing and adjusting weight parameters can further improve the matching accuracy of user demand intentions.

[0119] Embodiment 4:

[0120] S4. Data interaction and feedback: The cloud background transmits the analysis results back to the customer-specified interface or the user-side interface according to the user's request, and realizes data interaction and operation guidance based on the AI model.

[0121] Furthermore, the steps to implement data interaction include:

[0122] 1) Build a modular system architecture and select the core components of the ASR module. Strengthen the voice data through noise injection and accent enhancement, and improve the recognition accuracy based on context-aware ASR optimization through domain adaptation technology, expressed as: Where is the domain loss term. Use GST to achieve dynamic intonation adjustment, and embed the AI voice model in the user-side interface;

[0123] 2) The user-side interface includes the mobile APP, in-vehicle all-in-one machine, and streaming media rearview mirror. Access the diagnostic report based on the APP module and the program web page, and the AI voice model explains in real time according to the diagnostic report.

[0124] Specifically, the modular architecture design can improve the flexibility and maintainability of the system. The modular system architecture includes independent modules such as ASR, TTS, and NLP, supports hot plugging and independent upgrading of components. The ASR module selection can be compatible with open-source and commercial solutions, adapting to different cost and performance requirements. Achieve loose coupling between modules through a microservices architecture, and use API standardization to reduce the integration complexity.

[0125] The beneficial effects of the present invention are as follows: The method for collecting and processing automotive data combines automotive technical logic with AI algorithms based on automotive industry big data, combines traditional vehicle condition threshold judgment rules with the strong non-linear fitting ability of XGBoost, improves the real-time performance and reliability of diagnosis, adopts XGBoost-Boruta hybrid feature screening, improves the model interpretability and generalization ability compared with the traditional Dropout method, realizes the feature selection optimization strategy, and through incremental training and low-latency architecture, based on the distributed data acquisition and calculation framework of high-speed Internet of Things, supports the dynamic update of the vehicle-side data acquisition model sent by the background and the millisecond-level response of background calculation.

[0126] Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for collecting and processing automotive data, characterized in that It includes the following implementation steps: S1. Data collection: By installing a vehicle networking module behind the standard OBD diagnostic port of the vehicle, collect the vehicle bus data stream and upload it to the cloud background; S2. Data analysis: The cloud background analyzes and processes the vehicle networking data stream based on an algorithm model, and deeply analyzes the data in combination with the new content of the industry reference database and the industry knowledge base; S3. Weight adjustment: Develop potential demands based on user behavior, actively analyze user intentions and adjust data weights; S4. Data interaction feedback: The cloud background transmits the analysis results back to the customer-specified interface or the user terminal interface according to the user request, and realizes data interaction and operation guidance based on the AI model.

2. The method for collecting and processing automotive data according to claim 1, characterized in that, The vehicle connection module is adapted to the standard OBD diagnostic port of the vehicle. It can have an independent Internet of Things module directly connected to the background, or it can be connected to the cloud background data after being connected to the in-vehicle streaming media mirror or the user's mobile phone through Bluetooth communication. The specific steps of data collection include: 1) Connect the vehicle connection module to the vehicle-end OBD diagnostic port, and the cloud background dynamically issues data interaction instructions according to the vehicle model and scenario function requirements; 2) The OBD adapter in the vehicle connection module converts the data interaction instructions into vehicle interaction instructions to obtain vehicle bus information. The OBD adapter also converts the vehicle bus information into a vehicle networking data stream; 3) The vehicle networking device uploads the vehicle networking data stream to the cloud background through the Internet of Things, Bluetooth, and other wired connections; The cloud background for data aggregation and processing accesses the user client through the APP software module or the API interface, and is used to output the obtained vehicle networking data stream and the data diagnosis report to the user. Define the data collection range and collection frequency according to the actual usage scenario requirements and vehicle model support conditions. The highest data collection frequency is 20HZ.

3. A method for collecting and processing automotive data according to claim 1, characterized in that, The steps for the cloud background of vehicle data aggregation and processing to analyze and process the vehicle networking data stream based on an algorithm model include: 1) The cloud background parses and self-learns the obtained vehicle networking data stream; The data collected in the cloud database all carry hexadecimal numbers with timestamps. The cloud background automatically analyzes and processes them according to the requirements, including: a. Analyzing the environmental protection data related to vehicle emissions; b. Analyzing the data related to vehicle failures; c. Parsing the data of the performance of the whole vehicle and key components in combination with the vehicle brand and model characteristics; 2) Based on the forward diagnosis algorithm and the correlation algorithm of XGBoost, define the monitoring fields and thresholds according to the scenario requirements, and call the relevant calculation models when the preset conditions are met; 3) Integrate the results of multiple different calculation models, and the AI integrates and forms a report and outputs it; The steps for deeply analyzing data in combination with the new content of the industry reference database and the industry knowledge base include: 1) Connect to the government open platform, third-party databases, and industry research reports to access the industry reference database, write the knowledge base rule engine according to the new rules and update the domain ontology model, and add entity relationships in the knowledge graph; 2) Map the industry database and the knowledge base to achieve heterogeneous data fusion; 3) Extract domain features from CAN bus data in combination with the knowledge base, use the Apriori algorithm to mine association rules, perform rule constraint learning based on physical models and data-driven approaches, and construct a knowledge-enhanced hybrid model; 4) Establish an online learning mechanism to detect network attack patterns in real time and achieve incremental knowledge updates.

4. A method for collecting and processing automotive data according to claim 3, characterized in that The forward diagnostic algorithm based on data flow analyzes data, and the specific steps include: 1) Obtain the timing data of the whole vehicle and the engine, or the control logic of the electric drive system in real time, including parameters such as engine speed, temperature, or drive motor speed, power battery voltage and temperature, and transmit them with low latency based on high-speed Internet of Things communication technology to collect data streams in real time; 2) Use a state machine model to preset control logic thresholds to achieve control logic anomaly detection, including: a. Based on engine operating conditions judgment: if (speed > threshold and oil pressure < threshold), then trigger a diagnostic signal; b. Verify the robustness of the logic rules based on the hardware device on the HIL test platform; 3) Output the fault level based on the diagnostic rule base, combine with the finite state machine FSM to achieve multi-condition joint judgment, and generate a fault and special situation warning mechanism for real-time diagnostic decision-making, including: a. Determine all possible fault states and input conditions to be processed to clarify the diagnostic requirements and conditions, identify and list all condition combinations that need to be jointly judged, and define state behaviors by enumerating all states; b. Construct the transition relationship by defining possible transition paths and corresponding condition combinations for each state, describe the state transition logic using a transition table or UML state diagram, use logical operators to combine conditions to achieve multi-condition processing, and if multiple transition conditions are satisfied simultaneously, define priorities or adjust conditions; c. Represent the current state variable using enumeration or constants, trigger state checks when conditions change through event-driven, detect conditions regularly according to the polling mechanism, and construct the finite state machine FSM structure.

5. A method for collecting and processing automotive data according to claim 3, characterized in that, The correlation algorithm based on XGBoost analyzes data, and the specific steps include: 1) Input vehicle accident data as the first period and vehicle operation data as the second period based on the obtained Internet of Vehicles data stream, and perform data preprocessing operations, including missing value filling, data standardization, and feature alignment; 2) Adopt the XGBoost-Boruta hybrid strategy, perform preliminary screening of feature importance based on the Boruta algorithm, optimize feature selection and eliminate redundant variables, and construct an XGBoost model, where the gain calculation formula of XGBoost is expressed as: where G and H are the first / second derivatives of the loss function, and λ and γ are regularization parameters; 3) Construct the objective function, expressed as: Among them Used to control the complexity of the tree, train the XGBoost model based on the incremental training strategy, retain the prediction results of the previous t - 1 rounds in each round, and add a function (f t ) to optimize the residuals; 4) Verify the model accuracy based on the 1DCNN comparison method, and test the robustness on the HIL platform in combination with real-time vehicle data. The accuracy can reach 95% in the reclassification scenario, and it supports multi-label output of fault types.

6. The method for collecting and processing vehicle data according to claim 1, wherein The specific steps to analyze user needs and intentions and achieve data weight adjustment are: 1) Analyze the explicit and active demand data of users based on LP keyword extraction and conversation tree analysis and processing, and adjust the data weight range to 0.8 - 1.0; analyze the implicit and potential demand of users based on LSTM behavior sequence modeling and processing, and adjust the data weight range to 0.5 - 0.7; Analyze the passive-triggered demand of users based on collaborative filtering similarity matching and processing, and adjust the data weight range to 0.3 - 0.5; 2) Parse the text intention according to the BERT model, calculate the page area attention based on the eye movement tracking data, and construct a Markov model of the user operation chain to identify and analyze the user behavior time sequence; 3) Dynamically adjust the weight according to the weight calculation framework, expressed as: W i = α·I explicit + β·I implicit + γ·R context where α is the explicit intention factor, β is the implicit intention factor, and γ is the environmental factor.

7. A method for collecting and processing vehicle data according to claim 1, characterized in that, The steps to achieve data interaction include: 1) Build a modular system architecture and select the core components of the ASR module. Strengthen the speech data through noise injection and accent enhancement, and improve the recognition accuracy based on context-aware ASR optimization through domain adaptation technology, expressed as: Among them is the domain loss term, use GST to achieve intonation dynamic adjustment, and embed the AI speech model in the user interface; 2) The user-side interface includes a mobile APP, an in-vehicle all-in-one machine, and a streaming media rearview mirror. Access the diagnostic report based on the APP module and the program web page, and the AI voice model explains in real time according to the diagnostic report.

Images