A commercial vehicle driver safety scoring method based on CAN data

Abstract

The application discloses a kind of based on CAN data's commercial vehicle driver safety score method, comprising the following steps: S1, source data is uploaded to platform in the form of stream data;S2, real-time feature extraction is carried out in unit of trip, then first weighted average aggregation is carried out in unit of day to obtain feature record;S3, score of each driver per day is obtained using entropy weight method;S4, platform interface end obtains score according to time period, and carries out second weighted average aggregation, obtains the score of the driver in the time period;S5, platform front end shows according to the result data returned by interface end.The method of the application obtains the score of a certain driver in a time period by the combination of twice weighted average aggregation and entropy weight method, so as to overcome the defect that single entropy weight method in prior art cannot dynamically comprehensively evaluate the object in a time period.

Description

Technical Field

[0001] This invention relates to the field of vehicle networking data application technology, and in particular to a method for scoring the safety of commercial vehicle drivers based on CAN data. Background Technology

[0002] With the rapid development of the logistics industry and the gradual maturation of vehicle networking technology, the analysis and application of vehicle driving data has begun to receive attention. Research on the safety evaluation of commercial vehicle driving behavior based on vehicle networking data helps to monitor and regulate drivers' safe driving behavior, thereby improving traffic safety. Currently, the data for driving behavior safety evaluation mainly comes from ADAS, DSM, and edge hardware systems installed on vehicles, primarily GPS data. The evaluation method mainly uses the entropy weight method. However, the data collection intervals of these data sources are large, resulting in unreliable evaluation results. Furthermore, the entropy weight method is a static method that comprehensively evaluates all objects at once, with weights fixed after a single calculation. In reality, the evaluation objects may change over time, and the entropy value calculated for the same indicator may differ at different times. The proportion of information entropy redundancy to total information entropy redundancy may also vary, meaning the weights may differ at different times. The existing entropy weight method can only generate fixed weights for each feature of the current static data table at once, and cannot perform dynamic comprehensive evaluation of dynamic data tables over a period of time. Summary of the Invention

[0003] To address the aforementioned issues, this invention provides a method for scoring the safety of commercial vehicle drivers based on CAN data.

[0004] The present invention adopts the following technical solution:

[0005] A method for scoring the safety of commercial vehicle drivers based on CAN data includes the following steps:

[0006] S1. Use the on-board terminal data of commercial vehicles as the source data and upload it to the platform in the form of streaming data;

[0007] S2. Perform stream processing on the source data, extract features in real time on a trip-by-trip basis, and then perform a first weighted average aggregation on a daily basis to obtain feature records. Each driver corresponds to one feature record on the same day, and the feature records of all drivers form the feature space of the driver group on that day.

[0008] S3. Calculate the score for each driver per day using the entropy weight method on the feature space, and store the feature space and the score in the platform database;

[0009] S4. The platform interface obtains the scores at preset time intervals and performs a second weighted average aggregation of all scores for each driver within that time interval to obtain the driver's rating within that time interval.

[0010] S5. The platform front-end displays the result data returned by the interface.

[0011] Furthermore, the source data includes any one or more of CAN data, ADAS device data, Terminal data, and DSM data.

[0012] Furthermore, the stream processing employs a Kafka message queue.

[0013] Furthermore, each of the aforementioned feature records contains all the features of the corresponding driver and their corresponding feature values.

[0014] Furthermore, in the first weighted average aggregation, the higher the feature value, the higher its corresponding weight; in the second weighted average aggregation, the lower the score, the higher its corresponding weight.

[0015] Furthermore, the time period mentioned in step S4 can be set according to the user's needs.

[0016] Furthermore, step S3 also includes dividing the features into several feature groups according to the nature of the features, and then using the entropy weight method to calculate the feature group score for each driver per day.

[0017] Furthermore, the characteristic groups include a focus group, a risk awareness group, a vigilance group, a clear-headed group, and a stability group.

[0018] Furthermore, step S4 also includes the platform interface acquiring the feature group scores at preset time intervals according to a preset time period, and performing a second weighted average aggregation of all feature group scores for each driver within that time period to obtain the driver's total feature group scores within that time period.

[0019] Furthermore, the CAN data includes any one or more of the following: engine speed, acceleration, instantaneous fuel consumption, and braking frequency per kilometer; the ADAS device data includes any one or more of the following: lane departure warning, close proximity warning, and collision warning; the Terminal data includes any one or more of the following: driver fatigue warning and speeding warning; and the DSM data includes any one or more of the following: smoking warning, distraction warning, physiological fatigue warning, phone call warning, dangerous driving warning, and coasting in neutral on a long hill warning.

[0020] By adopting the above technical solution, the present invention has the following advantages compared with the prior art:

[0021] 1. This invention employs two weighted average aggregations throughout the safety scoring process. The first weighted average aggregation aggregates the features of a trip to the features of a day. This method is simple and effective, not only interpreting the meaning of features from a business perspective (i.e., highlighting the numerical values ​​of dangerous features by using weights) but also maintaining the monotonicity of features and degree of danger. The second weighted average aggregation calculates a weighted average of the scores of each driver on different days within a certain period to obtain the score for each driver during that period. Since each driver's departure date is different, the group of drivers on duty each day is different. The first weighted average aggregation is equivalent to sampling the driver group each day, while the second weighted average aggregation is equivalent to summing the number of times a driver is sampled over a period of time. The number of times each driver is sampled only has statistical significance when it reaches a certain level. Therefore, by combining the two weighted average aggregations with the entropy weight method, the score of a driver within a certain period of time can be obtained, thus overcoming the deficiency of the existing technology that cannot dynamically and comprehensively evaluate dynamic data tables within a certain period of time by relying solely on the entropy weight method.

[0022] 2. In addition, the present invention uses stream processing to extract features of source data in real time, thereby achieving dynamic comprehensive scoring of commercial vehicle drivers. Attached Figure Description

[0023] Figure 1 This is a flowchart of the method of the present invention;

[0024] Figure 2 This is a diagram showing the feature group scoring results in the embodiment. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0026] Example

[0027] like Figure 1 As shown, a method for scoring the safety of commercial vehicle drivers based on CAN data includes the following steps:

[0028] S1. The on-board terminal data of commercial vehicles is used as the source data and uploaded to the platform in the form of streaming data; the source data includes CAN data, ADAS device data, Term terminal data and DSM data.

[0029] This embodiment introduces CAN data. Since the data sampling interval of CAN data (about 1 second) is much smaller than that of GPS data, the characterization results will be more detailed, and the features extracted from data with the same meaning (such as vehicle speed) and the final analysis results will be more reliable.

[0030] S2. The source data undergoes stream processing using a Kafka message queue. Features are extracted in real-time on a per-trip basis, and then the extracted feature values ​​are aggregated using a first weighted average on a daily basis to obtain feature records. Each driver corresponds to one feature record for that day, and the feature records of all drivers form the feature space of the driver group for that day. Each feature record contains all the features of the corresponding driver and their corresponding feature values. During the first weighted average aggregation, the higher the feature value, the higher its corresponding weight; during the second weighted average aggregation, the lower the score, the higher its corresponding weight.

[0031] Since a driver may make more than one trip a day, features are extracted on a trip-by-trip basis, and then the trip features are aggregated to the day. After aggregation to the day, each driver corresponds to only one feature record. Each feature record contains multiple features and their corresponding feature values, such as the standard deviation of the weighted average acceleration greater than 0 per trip, the weighted average number of braking times per kilometer per trip, etc.

[0032] Specifically, the CAN data includes engine speed, acceleration, instantaneous fuel consumption, and braking frequency per kilometer. The features extracted from the CAN data are: the standard deviation of the first-order difference of the weighted average engine speed per trip that is greater than 0; the standard deviation of the weighted average acceleration per trip that is greater than 0; the standard deviation of the first-order difference of the weighted average engine speed per trip that is less than 0; the standard deviation of the weighted average acceleration per trip that is less than 0; the standard deviation of the first-order difference of the weighted average instantaneous fuel consumption per trip; the standard deviation of the first-order difference of the weighted average throttle position per trip; and the weighted average number of braking frequencies per kilometer per trip. The ADAS device data includes lane departure warning, close proximity warning, and collision warning. The features extracted from the ADAS device data are: the weighted average number of lane departure warnings, close proximity warnings, and collision warnings per trip per kilometer; the term terminal data includes fatigue driving and speeding warnings, and the features extracted from the term terminal data are: the weighted average alarm duration per trip per kilometer for fatigue driving and speeding warnings; the DSM data includes smoking, distraction alerts, physiological fatigue warnings, phone calls, dangerous driving warnings, and coasting in neutral on long slopes warnings, and the features extracted from the DSM data are: the weighted average number of alarms per kilometer for smoking, distraction alerts, physiological fatigue warnings, and phone calls.

[0033] The method of aggregating from trips to days uses a weighted average method. The idea behind the weight calculation is that the larger the feature value, the greater the weight (since the features in this embodiment are all inverse features, and the obtained score refers to the safety score, that is, the higher the score, the better. Therefore, the larger the feature, the lower the score. So, in order to highlight the risk, the larger the feature value, the greater its weight should be). The weight calculation method is to divide the corresponding feature value of the same feature of a driver in different trips by the sum of all feature values ​​of that driver in that trip. After aggregation to the day, a table of data for each driver will be formed. All features form the feature space for evaluating the driver group. In this way, all feature values ​​of each driver may be different every day, resulting in different feature spaces.

[0034] S3. Calculate the daily score for each driver using the entropy weight method on the feature space, and store the feature space and the score in the platform database. The daily score for all drivers can be obtained by using the entropy weight method on the feature space for each day. Therefore, the entropy weight method weights will be different for different days. In addition, since this is a real-time calculation of streaming data, the score for the day is also dynamically updated. The dynamic update for the day is because some trips have not yet ended, but the data before yesterday is fixed, so the score is also fixed.

[0035] Step S3 further includes dividing the features into several feature groups based on the nature of the features, such as... Figure 2 The radar chart shown has dimensional feature groups, including focus group, risk awareness group, vigilance group, alertness group, and stability group. Then, the entropy weight method is used to calculate the feature group score for each driver for each day.

[0036] Similarly, by applying the entropy weight method to the feature groups of all drivers for each day, we can obtain the feature group scores of all drivers for each day, grouped according to different dimensions.

[0037] S4. The platform interface obtains the score at regular intervals according to the user's preset time period. The time period can be selected and set according to the user's needs, such as any one of 3 days, 5 days, 7 days and 30 days. In this embodiment, the time period is 7 days. A second weighted average aggregation is performed on all the scores of each driver within the time period to obtain the driver's score within the time period.

[0038] On the interface side, all scores for each driver are aggregated. The aggregation principle is still that the more dangerous the driver, the higher the weight should be. The higher the score obtained on the interface side, the safer the driver is, and the lower the score, the more dangerous the driver is. The weight calculation method is to divide each score by the total score to obtain the weight of the same number of scores. Then, the larger the weight, the lower the score is weighted and the weighted average is calculated.

[0039] Step S4 also includes the platform interface acquiring the feature group scores at preset time intervals, and performing a second weighted average aggregation of all feature group scores for each driver within the time interval to obtain the driver's feature group scores within the time interval.

[0040] S5. Display the results data returned by the interface on the platform front end.

[0041] Similarly, a second weighted average aggregation is performed based on the feature group scores. The higher the feature group score obtained from the interface, the safer it is; the lower the score, the more dangerous it is. The specific weight calculation method is as follows: for each driver in each feature group, all scores for different dates are divided by the sum of the driver's scores for all dates. This yields the number of weights corresponding to the driver's date. These weights and scores are then sorted in different directions to ensure that smaller weights correspond to larger scores, and vice versa. Finally, a weighted sum is performed to obtain the driver's score for that feature group. This results in... Figure 2 The image shows the feature group scores for each driver over a period of time after being categorized according to different feature groups.

[0042] The above calculates the score for each driver in the radar chart dimension feature group. To calculate a finer-grained expanded dimension below the radar chart dimension, an additional inverse normalization step (inverse normalization for all driver scores) is required, as shown in the formula below:

[0043] (max-s) / (max-min+0.00000001)*100, where max represents the highest score among all drivers, min represents the lowest score among all drivers, and s represents the score of a particular driver.

[0044] Step S4 uses a second weighted average aggregation to ensure that each driver has only one score within a given time period (e.g., 7 days) before comparisons between drivers can be made. The reason for using a weighted average instead of a simple average is to highlight low scores (highlighting potential risks) by utilizing the weights. The lower the score, the higher the weight should be. The weight calculation method is simple and efficient: the ratio of a driver's total scores to the sum of all scores. A higher ratio assigns a lower weight to a lower score. Since each driver's work dates are different, the driver group varies daily. Each day is essentially a sample of the drivers; only a portion of the total drivers work each day. Therefore, the daily evaluation only considers this portion of drivers. The number of scores obtained by the interface for a particular driver represents the number of times that driver was sampled during that period. This sampling frequency needs to be sufficiently high to be statistically significant, representing the average level of multiple evaluations performed by that driver over a period of time.

[0045] S5. The platform front-end displays the result data returned by the interface.

[0046] To better understand the technical solution, this embodiment explains the entropy method and its calculation method.

[0047] 1. Introduction to the Entropy Method

[0048] The entropy method is an objective weighting method that determines the weight of each indicator based on the amount of information provided by its observed values; the greater the information content of an indicator, the more important it is. Given n samples and m evaluation indicators, the original indicator data matrix is ​​formed as follows:

[0049] X = (x ij ) nm (i=1,2,…,n; j=1,2,…,m)

[0050] Entropy is a measure of uncertainty. The greater the amount of information, the lower the uncertainty and the lower the entropy; conversely, the less information, the greater the uncertainty and the higher the entropy. Based on the characteristics of entropy, we can use entropy values ​​to determine the randomness and disorder of an event, and we can also use entropy to determine the dispersion of a certain indicator. The greater the dispersion of an indicator, the greater the amount of information, the lower the entropy, and the greater the influence of that indicator on the overall evaluation. Therefore, based on the information entropy redundancy of each indicator, we can calculate the weight of each indicator, providing a basis for multi-indicator comprehensive evaluation.

[0051] 2. Calculation steps (1) Index normalization processing

[0052] Because the units of measurement for various indicators are not standardized, they must be normalized before calculating the composite index. Since positive and negative indicators have different meanings, different methods are used for data normalization.

[0053] Positive indicators:

[0054]

[0055] Where δ is an infinitesimal.

[0056] Negative indicators:

[0057]

[0058] Then x ij Let j be the j-th index value of the i-th sample, where i = (1, 2, ..., n); j = (1, 2, ..., m);

[0059] (2) Calculate the proportion of each sample value under each indicator.

[0060]

[0061] (3) Calculate the entropy value of the j-th index.

[0062]

[0063] Where, n∈N + It can be proven that 0 ≤ e j ≤log2n.

[0064] (4) Since the lower the entropy of an indicator, the greater its weight should be, an information entropy redundancy is defined:

[0065] d j =log2n-e j

[0066] (5) Calculate the weights:

[0067]

[0068] (6) Calculate the score for each sample:

[0069] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for scoring the safety of commercial vehicle drivers based on CAN data, characterized in that: Includes the following steps: S1. Using the vehicle-mounted terminal data of commercial vehicles as source data, the source data includes any one or more of CAN data, ADAS device data, Term terminal data and DSM data, and uploads it to the platform in the form of streaming data. S2. The source data is stream-processed using a Kafka message queue, and features are extracted in real time on a trip-by-trip basis. Then, the extracted feature values ​​are aggregated by weighted average on a daily basis to obtain feature records. Each driver corresponds to one feature record on the same day, and the feature records of all drivers form the feature space of the driver group on that day. S3. Calculate the score for each driver per day using the entropy weight method on the feature space, and store the feature space and the score in the platform database; S4. The platform interface obtains the scores at preset time intervals and performs a second weighted average aggregation of all scores for each driver within that time interval to obtain the driver's rating within that time interval. S5. The platform front-end displays the result data returned by the interface.

2. The commercial vehicle driver safety scoring method based on CAN data as described in claim 1, characterized in that: Each of the aforementioned feature records contains all the features of the corresponding driver and their corresponding feature values.

3. The commercial vehicle driver safety scoring method based on CAN data as described in claim 2, characterized in that: In the first weighted average aggregation, the higher the feature value, the higher its corresponding weight; in the second weighted average aggregation, the lower the score, the higher its corresponding weight.

4. The commercial vehicle driver safety scoring method based on CAN data as described in claim 3, characterized in that: The time period mentioned in step S4 is set according to the user's needs.

5. The commercial vehicle driver safety scoring method based on CAN data as described in claim 4, characterized in that: Step S3 also includes dividing the features into several feature groups according to the nature of the features, and then using the entropy weight method to calculate the feature group score for each driver per day.

6. The commercial vehicle driver safety scoring method based on CAN data as described in claim 5, characterized in that: The characteristic groups include the focus group, risk awareness group, vigilance group, alertness group, and stability group.

7. The commercial vehicle driver safety scoring method based on CAN data as described in claim 6, characterized in that: Step S4 also includes the platform interface acquiring the feature group scores at preset time intervals according to a time period, and performing a second weighted average aggregation of all feature group scores for each driver within that time period to obtain the driver's total feature group scores within that time period.

8. A commercial vehicle driver safety scoring method based on CAN data as described in any one of claims 1-7, characterized in that: The CAN data includes any one or more of the following: engine speed, acceleration, instantaneous fuel consumption, and braking frequency per kilometer; the ADAS device data includes any one or more of the following: lane departure warning, close proximity warning, and collision warning; the Terminal data includes any one or more of the following: driver fatigue warning and speeding warning; the DSM data includes any one or more of the following: smoking warning, distraction warning, physiological fatigue warning, phone call warning, dangerous driving warning, and coasting in neutral on a long hill warning.

Images