A near real-time cleaning data execution method based on data collected by vehicle-mounted terminals

Abstract

This invention provides a near real-time data cleaning method based on data collected from vehicle-mounted terminals. The method includes a vehicle-mounted terminal data collection module, a multi-source data simple processing module, a multi-source data complex replacement module, a driving behavior calculation module, and an offline query OLAP data analysis module. By using multiple filtering methods such as dynamic acquisition of basic vehicle parameters, configurable preset thresholds, baseline caching, and linear growth fitting change filtering, various types of data acquired from the vehicle-mounted terminal are filtered multiple times. Redundant, invalid, and erroneous data are removed before data replacement and effective filling. Compared to previous vehicle data processing methods, this method is more logically complex. Because it only processes data from the earlier time period, it has the advantage of low load. Processing nodes can quickly complete tasks, resulting in significantly improved data processing speed, significantly improved accuracy, and a higher variety of processed data.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, specifically to a near real-time cleaning data execution method based on data collected by an on-board terminal. Background Technology

[0002] During vehicle operation, the onboard terminal uploads massive amounts of data, including a large amount of invalid and erroneous data that is ineffective for calculating and statistically analyzing vehicle driving events. This includes issues such as inconsistent data timing, missed data reports, data anomalies, data jumps, and data loss during transmission. These problems have already caused irreparable errors with the current computing solutions, severely impacting the company's related business operations.

[0003] Currently, driving behavior data is calculated offline using a T-1 day model. All data is read during the calculation, including data transformations and outliers. Filtering measures such as the median are also used during the calculation. This technical approach has the following drawbacks:

[0004] ① The calculation of driving behavior is delayed and slow;

[0005] ② The calculation process may result in the failure to record intermediate states due to transformations and anomalies, and the calculation results may not be consistent each time;

[0006] ③ The filtering measures used are too simplistic, resulting in data filtering failures and calculation results that exceed the normal range. Summary of the Invention

[0007] Therefore, the present invention provides a near real-time data cleaning execution method based on data collected by vehicle-mounted terminals, which can overcome the defects of existing technologies such as delayed calculation of driving behavior, failure to record intermediate states due to processing transformation and anomalies during the calculation process, inconsistent calculation results, and relatively simple filtering measures, resulting in data filtering failure and calculation results exceeding the normal range.

[0008] To address the aforementioned problems, this invention provides a near real-time data cleaning execution method based on data collected by an onboard terminal, comprising the following steps:

[0009] Step 1: Using the vehicle terminal data acquisition module, collect multi-source raw data of the vehicle during driving. The multi-source raw data is divided into packets according to time periods and stored to form several data packets of different time periods. The data packets are cleaned and fused to obtain a base data packet.

[0010] Step 2: In the multi-source data simple processing module, the theoretical range value of the reference data packet is determined, and the theoretical range value is compared with the collected value in the reference data packet. The collected value is retained or removed to obtain a standardized format data packet.

[0011] Step 3: In the complex substitute module for multi-source data, different source data within the format data packet are associated and merged, and reasonable time leveling is set. The different source data are then removed, replaced, and filled to obtain multi-dimensional data of the vehicle at the same time.

[0012] Step 4: Within the driving behavior calculation module, the multi-dimensional data obtained in Step 3 is used to calculate vehicle driving behavior, and the daily vehicle driving behavior data is calculated and stored in the corresponding files and database.

[0013] Step 5: Call offline query OLAP data analysis, clean the vehicle reported data and write it into the Hive warehouse in plaintext in real time, and use HSQL statements to analyze the historical data.

[0014] In some implementations, the multi-source raw data includes CAN, GPS, and basic vehicle parameters stored using Hadoop files.

[0015] In some implementations, the data packet execution process in step 1 includes the following steps:

[0016] Step 1.1 When the data packet is executed, the data packets are sorted according to their time sequence, and the executed data packets and the computer system time are obtained.

[0017] Step 1.2 Determine if the time of the data packet being processed is less than 60 minutes before the system time, and if the data packet being processed is the earliest data packet among the unprocessed data packets of the day.

[0018] Step 1.3: The baseline data obtained after cleaning the previous data packet is denoted as d1, and the baseline data of the currently executed data packet is denoted as d2. When d1 <= d2, the cleaning of the currently executed data packet ends, and the next data packet is executed. When d1 > d2, the currently executed data packet is fused with the previous data packet, and the baseline data obtained after the fused data packet is denoted as d3. All data packets of the day are executed sequentially until the execution is completed.

[0019] In some implementations, the data packets collected daily are divided into segments every half hour.

[0020] In some implementations, the execution process of the multi-source data simple processing module in step 2 includes the following steps:

[0021] Step 2.1 Confirmation of the theoretical range value includes using the data collected and reported by 1 million vehicles on a certain platform over the past three years as the basic data, and using the four types of data reported by a single vehicle in the past year—cumulative mileage, cumulative fuel consumption, cumulative running time, and vehicle speed—to calculate the growth coefficient value 'a' for mileage, fuel consumption, running time, and vehicle speed respectively using linear fitting. Then, using this coefficient 'a', the theoretical range value of the cumulative value within a given time period is calculated.

[0022] Step 2.2 Perform data removal or data retention on the executed baseline data packet.

[0023] In some implementations, the execution process of the multi-source data complex substitute module in step 3 includes the following steps:

[0024] Step 3.1 Data item anomaly judgment, judgment condition ①: adjacent previous data value <= current input data value <= adjacent next data value; condition ②: (current input data value - adjacent previous data value) / (current input data time - adjacent previous data value time) <= theoretical range of change rate; if at least one condition of the current input data is not met, the current input data is marked with a fill mark and is data to be filled.

[0025] Step 3.2 First, the data to be filled is filled for the first time using the same data type to obtain the data after the first filling and make it meet condition ①. Then, it is determined whether the data after the first filling meets condition ②. If it does, the filling ends; otherwise, the next filling command is executed.

[0026] Step 3.3 First, the data after the first filling is filled a second time by adding the change value of the adjacent previous data value and the latitude and longitude adjacent data value to obtain the data after the second filling. The data after the second filling satisfies condition ①. Then, it is determined whether the data after the second filling satisfies condition ②. If it does, the filling ends. If it does not, the next filling command is executed.

[0027] Step 3.4: Add the theoretical value to the adjacent previous data value after the second filling, and end the filling process. Where X N X represents the Nth data value. N-1 t represents the (N-1)th data value. N -t0 represents the time difference between the Nth data value and the 0th data value, where N is 1, 2, 3...n.

[0028] Step 3.5 completes the filling process and outputs the multi-dimensional data.

[0029] In some implementations, the baseline data packet output in step 1 and the multi-dimensional data output in step 3 are both based on time for fusion, and the fused data structure is "vehicle|time|pulse mileage|vehicle speed|latitude and longitude|", or "vehicle|basic parameters".

[0030] In some implementations, the driving behavior includes a baseline mileage, fuel consumption, and runtime.

[0031] This invention provides a near real-time data cleaning execution method based on data collected from vehicle-mounted terminals. It utilizes a triggered MapReduce computing framework to execute data within a specific time period in stages, prioritizing the cleaning of data as early as possible to ensure same-day data cleaning is completed. The MapReduce framework is suitable for parallel processing of large-scale datasets and is ideal for processing files persisted to disk. This includes joint calculations based on data time, data type, value, and basic vehicle parameters. The cleaned data results are appended to Hive, supporting ad-hoc queries using HSQL. Furthermore, this application employs multiple filtering methods, such as dynamic acquisition of basic vehicle parameters, configurable preset thresholds, baseline caching, and linear growth fitting variation filtering, to filter various types of data acquired from the vehicle-mounted terminal multiple times. Redundant, invalid, and erroneous data are removed before data replacement and effective filling. Compared to previous vehicle data processing methods that are more logically complex, this method has the advantage of low load due to processing only data from the earliest time periods. Processing nodes can quickly complete tasks, resulting in significantly improved data processing speed, accuracy, and a wider variety of processed data. This application also supports ad-hoc queries of historical vehicle data and OLAP offline data analysis services. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the data processing flow of the near real-time cleaning data execution method based on data collected by an in-vehicle terminal according to an embodiment of the present invention;

[0033] Figure 2 This is a schematic diagram of the multi-source data simple processing module of the near real-time data cleaning execution method based on data collected by vehicle terminal in an embodiment of the present invention (taking mileage calculation as an example);

[0034] Figure 3 This is a schematic diagram of the multi-source data complex replacement module processing flow of the near real-time data cleaning execution method based on data collected by vehicle terminal according to an embodiment of the present invention (taking mileage calculation as an example);

[0035] Figure 4 This is an example diagram of linear fitting for a near real-time cleaning data execution method based on data collected by an on-board terminal, according to an embodiment of the present invention. Detailed Implementation

[0036] See also Figures 1 to 4 As shown in the embodiment of the present invention, a near real-time cleaning data execution method based on data collected by an on-board terminal is provided, including the following steps:

[0037] Step 1: Using the vehicle terminal data acquisition module, collect multi-source raw data of the vehicle during driving. The multi-source raw data is divided into packets according to time periods and stored to form several data packets of different time periods. The data packets are cleaned and fused to obtain a base data packet.

[0038] Step 2: In the multi-source data simple processing module, the theoretical range value of the reference data packet is determined, and the theoretical range value is compared with the collected value in the reference data packet. The collected value is retained or removed to obtain a standardized format data packet.

[0039] Step 3: In the complex substitute module for multi-source data, different source data within the format data packet are associated and merged, and reasonable time leveling is set. The different source data are then removed, replaced, and filled to obtain multi-dimensional data of the vehicle at the same time.

[0040] Step 4: Within the driving behavior calculation module, the multi-dimensional data obtained in Step 3 is used to calculate vehicle driving behavior, and the daily vehicle driving behavior data is calculated and stored in the corresponding files and database.

[0041] Step 5: Call offline query OLAP data analysis, clean the vehicle reported data and write it into the Hive warehouse in plaintext in real time, and use HSQL statements to analyze the historical data.

[0042] In one specific embodiment, the multi-source raw data includes CAN, GPS, and basic vehicle parameters stored using HADOOP files.

[0043] In a specific embodiment, the data packet execution process in step 1 includes the following steps:

[0044] Step 1.1 When the data packet is executed, the data packets are sorted according to their time sequence, and the executed data packets and the computer system time are obtained.

[0045] Step 1.2 Determine if the time of the data packet being processed is less than 60 minutes before the system time, and if the data packet being processed is the earliest data packet among the unprocessed data packets of the day.

[0046] Step 1.3: The baseline data obtained after cleaning the previous data packet is denoted as d1, and the baseline data of the currently executed data packet is denoted as d2. When d1 <= d2, the cleaning of the currently executed data packet ends, and the next data packet is executed. When d1 > d2, the currently executed data packet is fused with the previous data packet, and the baseline data obtained after the fused data packet is denoted as d3. All data packets of the day are executed sequentially until the execution is completed.

[0047] In one specific embodiment, the data packets collected daily are divided into segments every half hour.

[0048] In a specific embodiment, the execution process of the multi-source data simple processing module in step 2 includes the following steps:

[0049] Step 2.1 Confirmation of the theoretical range value includes using the data collected and reported by 1 million vehicles on a certain platform over the past three years as the basic data, and using the four types of data reported by a single vehicle in the past year—cumulative mileage, cumulative fuel consumption, cumulative running time, and vehicle speed—to calculate the growth coefficient value 'a' for mileage, fuel consumption, running time, and vehicle speed respectively using linear fitting. Then, using this coefficient 'a', the theoretical range value of the cumulative value within a given time period is calculated.

[0050] Step 2.2 Perform data removal or data retention on the executed baseline data packet.

[0051] Specifically, taking the cumulative mileage as an example, assuming the cumulative mileage range is 0 < m <= 1800000 km, the range value can be predicted through the above fitting to correct m and obtain m'. Then, it is judged whether the reported data (i.e., m) and m' are within the range value. For the confirmation of the theoretical vehicle speed range value, first input the vehicle file data. Using the current inherent vehicle speed calculation formula Ua = 0.377 * r * n / ig / io, where Ua is the current vehicle speed, r is the wheel radius, n is the rotational speed, ig is the transmission ratio of the gearbox, and io is the main reducer ratio (rear axle speed ratio), the rated maximum vehicle speed v1 is calculated. Then, according to the vehicle speed values reported by the single vehicle in the past year, the vehicle speeds are first sorted naturally and a discrete distribution is processed. Taking the lowest vehicle speed 0 as the center and recording it as 0%, and taking the central vehicle speed of the top 5% data of the reported maximum vehicle speed as 100%, the data outside the discrete rate outer circle is removed, and then a normal distribution is processed. The average value of the data in the top 20% part of the normal distribution graph is obtained as the theoretical vehicle speed v2. The method of taking v1 and v2 satisfies the following algorithm: when v1 = 0, take the value of v2 and record it as v; when the difference between v1 and v2 is close to 20% of the larger value, take the maximum value and record it as v, otherwise take the value of v2 and record it as v. The theoretical vehicle speed v is obtained from v1 and v2, and the theoretical vehicle speed v, the reported mileage data, and the reference value updated in the previous packet are compared. If the following two conditions are met, it is a normal value. These two conditions are respectively: Condition ①: mileage data >= reference value; Condition ②: the mileage difference between two adjacent mileage data (taking the absolute value of the subtraction of the previous value from the next value) <= time difference * theoretical vehicle speed v, otherwise this data is removed.

[0052] Specifically, the linear fitting process in step 2.1 is as follows:

[0053] Use sample data, that is, the data reported during a historical driving process, for linear fitting:

[0054] Vehicle ID | Data time | Cumulative mileage data

[0055] 2126082338 | 1684512000 | 15725860

[0056] 2126082338 | 1684512002 | 15725861

[0057] 2126082338 | 1684512004 | 15725862

[0058] 2126082338 | 1684512006 | 15725863

[0059] 2126082338 | 1684512008 | 15725864

[0060] 2126082338|1684512010|15725856

[0061] 2126082338|1684512012|15725866

[0062] 2126082338|1684512014|15725867

[0063] 2126082338|1684512016|15725868

[0064] 2126082338|1684512018|15725869

[0065] 2126082338|1684512020|15725870

[0066] 2126082338|1684512022|15725879

[0067] 2126082338|1684512024|15725872

[0068] 2126082338|1684512026|15725879

[0069] 2126082338|1684512028|15725874

[0070] 2126082338|1684512031|15725875

[0071] 2126082338|1684512033|15725876

[0072] 2126082338|1684512035|15725877

[0073] 2126082338|1684512037|15725878

[0074] 2126082338|1684512039|15725879

[0075] 2126082338|1684512041|15725880

[0076] 2126082338|1684512043|15725881

[0077] 2126082338|1684512045|15725886

[0078] 2126082338|1684512047|15725883

[0079] 2126082338|1684512048|15725884

[0080] 2126082338|1684512050|15725888

[0081] 2126082338|1684512052|15725890

[0082] 2126082338|1684512054|15725890

[0083] 2126082338|1684512056|15725894

[0084] 2126082338|1684512058|15725894

[0085] 2126082338|1684512061|15725890

[0086] 2126082338|1684512063|15725891

[0087] 2126082338|1684512065|15725892

[0088] 2126082338|1684512067|15725893

[0089] 2126082338|1684512069|15725894

[0090] 2126082338|1684512071|15725895

[0091] 2126082338|1684512073|15725899

[0092] 2126082338|1684512075|15725897

[0093] 2126082338|1684512077|15725897

[0094] 2126082338|1684512079|15725899

[0095] 2126082338|1684512081|15725901

[0096] 2126082338|1684512083|15725900

[0097] 2126082338|1684512085|15725902

[0098] 2126082338|1684512087|15725904

[0099] 2126082338|1684512089|15725904

[0100] 2126082338|1684512091|15725905

[0101] 2126082338|1684512093|15725906

[0102] 2126082338|1684512095|15725906

[0103] 2126082338|1684512097|15725905

[0104] 2126082338|1684512099|15725905

[0105] like Figure 4 As shown, by using time and cumulative mileage data to perform linear fitting, the growth coefficient value 'a' of the dashed line in the graph can be obtained, thus revealing the quantitative relationship of interdependence between variables. The dashed line is made as close as possible to all points, with a similar number of points above and below it. Least squares regression is then used to calculate the slope of the dashed line.

[0106] The formula for linearly finding 'a' is: a = nΣ(xy) – Σx*Σy*nΣ(x 2 )–(Σx) 2 n is the number of points.

[0107] The operation involves the following steps:

[0108] Step 1: For each (x, y) point, calculate x² and xy.

[0109] Step 2: All x, y, x 2 Adding x and y, we get Σx, Σy, Σx 2 and Σxy

[0110] Step 3: Calculate the slope a: a = nΣ(xy) – Σx*Σy*nΣ(x 2 )–(Σx) 2

[0111] The result 'a' of the linear fitting calculation using historical data can be used to calculate the theoretical cumulative value within a given time period, which can be used to effectively compare and verify the data to be reported in the future.

[0112] In a specific embodiment, the execution process of the multi-source data complex substitute module in step 3 includes the following steps:

[0113] Step 3.1 Data item anomaly judgment, judgment condition ①: adjacent previous data value <= current input data value <= adjacent next data value; condition ②: (current input data value - adjacent previous data value) / (current input data time - adjacent previous data value time) <= theoretical range of change rate; if at least one condition of the current input data is not met, the current input data is marked with a fill mark and is data to be filled.

[0114] Step 3.2 First, the data to be filled is filled for the first time using the same data type to obtain the data after the first filling and make it meet condition ①. Then, it is determined whether the data after the first filling meets condition ②. If it does, the filling ends; otherwise, the next filling command is executed.

[0115] Step 3.3 First, the data after the first filling is filled a second time by adding the change value of the adjacent previous data value and the latitude and longitude adjacent data value to obtain the data after the second filling. The data after the second filling satisfies condition ①. Then, it is determined whether the data after the second filling satisfies condition ②. If it does, the filling ends. If it does not, the next filling command is executed.

[0116] Step 3.4: Add the theoretical value to the adjacent previous data value after the second filling, and end the filling process. Where X N X represents the Nth data value. N-1 t represents the (N-1)th data value. N -t0 represents the time difference between the Nth data value and the 0th data value, where N is 1, 2, 3...n.

[0117] Step 3.5 completes the data filling and outputs the multi-dimensional data. Examples include: CAN cumulative mileage data "Vehicle|Time|Pulse Mileage", CAN cumulative fuel (gas) data "Vehicle|Time|Fuel (Gas)", CAN engine operation data "Vehicle|Time|Cumulative Running Time", CAN vehicle speed data "Vehicle|Time|Vehicle Speed", and GPS mileage data "Vehicle|Time|GPS Mileage", etc.

[0118] In a specific embodiment, the baseline data packet output in step 1 and the multi-dimensional data output in step 3 are both based on time for fusion, and the fused data structure is "vehicle|time|pulse mileage|vehicle speed|latitude and longitude|", or "vehicle|basic parameters".

[0119] Specifically, let's take mileage calculation as an example for a detailed explanation:

[0120] The rate of change range (v) is modified based on the theoretical range of vehicle speed in the simple cleaning process, and the default value is v = [0, 150]. The latitude and longitude change range is calculated based on the distance d1 and time difference t1 between two adjacent latitude and longitude points in the natural time sequence of the reported data. The distance difference d2 and d3, and the time difference t2 and t3 can be obtained between two adjacent upper and lower data points.

[0121] By comparing the range of vehicle speed change rate: whether the vehicle speed is within the range of v, if so, the condition is met; by comparing the range of latitude and longitude change: whether d1 / t1 is within the range of v, if so, the condition is met; adjacent upper and lower d2 and d3 respectively satisfy whether d2 / t2 and d3 / t3 are within the range of v. If none of the above are satisfied, v*t (the time difference between two adjacent time points in natural time order) = d, and using d plus the mileage data of the previous pulse that is adjacent in natural time order to replace the mileage of the current position. This mileage replacement ensures that the frequency is very regular after data cleaning, making it more accurate when calculating driving events.

[0122] In one specific embodiment, the driving behavior includes a baseline mileage, fuel consumption, and runtime.

[0123] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention. The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the protection scope of the present invention.

Claims

1. A near real-time data cleaning execution method based on data collected by an on-board terminal, characterized in that, Including the following steps: Step 1: Using the vehicle terminal data acquisition module, collect multi-source raw data of the vehicle during driving. The multi-source raw data is divided into packets according to time periods and stored to form several data packets of different time periods. The data packets are cleaned and fused to obtain a base data packet. Step 2: In the multi-source data simple processing module, the theoretical range value of the reference data packet is determined, and the theoretical range value is compared with the collected value in the reference data packet. The collected value is retained or removed to obtain a standardized format data packet. Step 3: In the multi-source data complex substitute module, different source data within the format data packet are associated and merged, and reasonable time leveling processing is set. The different source data are removed, replaced, and filled to obtain multi-dimensional data of the vehicle at the same time. The execution process of the multi-source data complex substitute module in step 3 includes the following steps: Step 3.1 Data item anomaly judgment, judgment condition ①: adjacent previous data value <= current input data value <= adjacent next data value; condition ②: (current input data value - adjacent previous data value) / (current input data time - adjacent previous data value time) <= theoretical range of change rate; if at least one condition of the current input data is not met, a fill mark is added to the current input data, and it is data to be filled; Step 3.2 First, the data to be filled is filled for the first time using the same data type to obtain the data after the first filling and make it meet condition ①. Then, it is determined whether the data after the first filling meets condition ②. If it meets the condition, the filling ends; otherwise, the next filling command is executed. Step 3.3 First, the data after the first filling is filled a second time by adding the change value of the adjacent previous data value and the latitude and longitude adjacent data value to obtain the data after the second filling. The data after the second filling satisfies condition ①. Then, it is determined whether the data after the second filling satisfies condition ②. If it does, the filling ends. If it does not, the next filling command is executed. Step 3.4: After the second filling, add the theoretical value to the adjacent previous data value and end the filling process. The theoretical value... ,in Represents the Nth data value. Represents the (N-1)th data value. This represents the time difference between the Nth data value and the 0th data value, where N is 1, 2, 3...n; Step 3.5 completes the data filling and outputs the multi-dimensional data; Step 4: Within the driving behavior calculation module, the multi-dimensional data obtained in Step 3 is used to calculate vehicle driving behavior, and the daily vehicle driving behavior data is calculated and stored in the corresponding files and databases. Step 5: Call offline query OLAP data analysis, clean the vehicle reported data and write it into the Hive warehouse in plaintext in real time, and use HSQL statements to analyze the historical data.

2. The near real-time cleaning data execution method based on data collected by an on-board terminal according to claim 1, characterized in that, The multi-source raw data includes CAN, GPS, and basic vehicle parameters stored using Hadoop files.

3. The near real-time cleaning data execution method based on data collected by an on-board terminal according to claim 2, characterized in that, The data packet execution process in step 1 includes the following steps: When the data packets are executed in step 1.1, they are sorted according to their time sequence, and the executed data packets and the computer system time are obtained. Step 1.2 Determine if the time of the data packet being processed is less than 60 minutes before the system time, and if the data packet being processed is the earliest data packet among the unprocessed data packets of the day; Step 1.3: The baseline data obtained after cleaning the previous data packet is denoted as d1, and the baseline data of the currently executed data packet is denoted as d2. When d1 <= d2, the cleaning of the currently executed data packet ends, and the next data packet is executed. When d1 > d2, the currently executed data packet is fused with the previous data packet, and the baseline data obtained after the fused data packet is denoted as d3. All data packets of the day are executed sequentially until all are completed.

4. The near real-time cleaning data execution method based on data collected by an on-board terminal according to claim 3, characterized in that, The data packets collected daily are divided into segments every half hour.

5. The near real-time cleaning data execution method based on data collected by an on-board terminal according to claim 1, characterized in that, The execution process of the multi-source data simple processing module in step 2 includes the following steps: Step 2.1 Confirmation of the theoretical range value includes using data collected and reported by 1 million vehicles over the past three years as the basic data, and using linear fitting to calculate the growth coefficient value 'a' for each of the four types of data reported by a single vehicle in the past year: cumulative mileage, cumulative fuel consumption, cumulative running time, and vehicle speed. Then, using this coefficient 'a', the theoretical range value of the cumulative value within a given time period is calculated. Step 2.2 Perform data removal or data retention on the executed baseline data packet.

6. The near real-time cleaning data execution method based on data collected by an on-board terminal according to claim 1, characterized in that, The baseline data packet output in step 1 and the multi-dimensional data output in step 3 are both based on time for fusion. The fused data structure is "vehicle|time|pulse mileage|vehicle speed|latitude and longitude|", or "vehicle|basic parameters".

7. The near real-time cleaning data execution method based on data collected by an on-board terminal according to claim 1, characterized in that, The driving behavior includes baseline mileage, fuel consumption, and runtime.

Images