A method for extending data of expected functional safety test of Internet of Vehicles

Abstract

The present invention provides a method for expanding data of expected functional safety tests of an Internet of Vehicles, and relates to the field of safety technology of the Internet of Vehicles. The present invention expands the data of emergencies and rare driving behaviors in an original data set, so that the total amount of data becomes larger. The method for expanding data in an expected functional safety test data set expands the data of emergencies and rare driving behaviors in the data set. Compared with normal driving behavior data, the proportion of emergencies and rare driving behavior data in the original data set is relatively small. After data expansion, the number of emergencies and rare driving behavior data increases, which can make the training model more sensitive to emergencies and rare driving behaviors.

Description

Technical Field

[0001] The present invention relates to the field of vehicle networking safety technology, and in particular to a method for expanding safety test data of expected functions of a vehicle networking. Background Art

[0002] In the field of Internet of Vehicles security or business scenarios, in order to solve the problem of difficulty in obtaining technical training data for the expected functional safety of Internet of Vehicles and the overall lack of expected functional safety test data sets, current automobile companies usually obtain data sets for expected functional safety tests, which are mostly data collected or trained by various research institutions, technology companies and schools in virtual driving scenarios.

[0003] A method for expanding vehicle network expected function safety test data proposed in the present invention is to expand the generated data to expand the expected function safety test data based on the existing expected function safety test data set, especially to strengthen the data expansion of emergencies and rare driving behaviors in the data set, so that the expected function safety development team can expand the data set and increase the sample ratio of emergencies and rare driving behaviors after collecting data from a specific area, thereby enhancing the comprehensiveness of the expected function safety test. Summary of the invention

[0004] In view of the above problems, the present invention proposes a method for expanding data of safety test of expected functions of Internet of Vehicles, which specifically comprises the following steps:

[0005] SA detects a call request of the Internet of Vehicles platform model. If the call request is detected, SB is executed. If the call request is not detected, the Internet of Vehicles platform continues to be monitored;

[0006] SB. If the call request is detected, determining whether the Internet of Vehicles platform is to further improve the trained model or train a new model;

[0007] SC. If a new model is to be trained, the process of training a new model will be started;

[0008] SD. If the trained model is to be further improved, the driving scene M1 corresponding to the model M to be trained is determined;

[0009] SE. Query whether there is a trained model X in the trained models corresponding to the driving scenario M1;

[0010] SF. If the trained model X exists, the trained model X is fed back to the Internet of Vehicles platform to allow the platform to confirm whether the model X can meet the requirements;

[0011] SG. If the model X can meet the platform requirements, directly call the model X and apply it;

[0012] SH. If the model X fails to meet the platform requirements, start the process of training a new model.

[0013] Furthermore, the step SA specifically includes the following steps:

[0014] The Internet of Vehicles platform model calling request is issued by the Internet of Vehicles server.

[0015] Furthermore, the step SB specifically includes the following steps:

[0016] The Internet of Vehicles server can only execute two types of processes: improving the trained model and training a new model.

[0017] Furthermore, the training new model process in step SC is:

[0018] S1. Use K-means algorithm to cluster emergency and rare driving behavior data;

[0019] S2. Perform sample expansion in each of the clusters and execute step S3;

[0020] S3. Use the Euclidean distance to calculate the K similar samples closest to xi;

[0021] S4. Perform linear difference among K similar samples;

[0022] S5. Add Gaussian white noise to each sample generated by interpolation to increase randomness;

[0023] S6. Using the distance of the cluster centers finally obtained in the K-means clustering process, filter the newly added sample points in different classes to form a sample set for completing data expansion;

[0024] S7. Perform model training based on the sample set that has completed data expansion to complete the training of the new model.

[0025] Furthermore, in S1, the K-means algorithm is used to cluster the emergency and rare driving behavior data to distinguish how many types of emergency and rare driving behaviors there are in the collected data. Since the basic data has labels, and the newly expanded data must be related to the data in the real world, the data is first classified, and then the newly expanded data can directly carry the labels. In the K-means clustering process, the choice of K determines the convergence speed of the calculation. Therefore, the value of K can be manually defined after manually browsing the data.

[0026] Furthermore, in S2, data are expanded in different clusters respectively. Since the core method of expanding data in this method is linear interpolation and Gaussian white noise error, it is necessary to ensure the consistency and integrity of the expanded data with the actual driving scene. Therefore, the emergency and rare driving behavior data are clustered first, and data expansion is performed in specific areas of data distribution.

[0027] Furthermore, in S3-S4, the Euclidean distance is used to calculate the K1 similar samples that are closest to xi. That is, in each cluster, K1 sample points with the closest distance are selected. Here x i Represents the data of the i-th emergency and rare driving behavior in a cluster. That is, the distance between any two points in the cluster is calculated, and then K1 points with the smallest distance are selected. Interpolation is performed among these K1 points. The core expression of the interpolation process is:

[0028]

[0029] where p ki The k in y is the kth point among the K1 points with the smallest distance, and i is the ith point in a cluster. k Indicates that among all the points in the cluster, the kth point with the smallest distance between any two points is calculated, x i is the point representing the i-th emergency situation and rare driving behavior in a cluster. ki For k With x i The number of z is determined by the scale to be expanded. rand(0,1) represents a random number between 0 and 1.

[0030] Furthermore, in S5, each expanded point is subjected to Gaussian white noise to enhance randomness. The Gaussian white noise expression is as follows:

[0031]

[0032] Where w is Gaussian white noise, that is, the error of Gaussian distribution. μ is the expected value of the distance between the K1 points with the smallest distance, and σ is the standard deviation of the distance between the K1 points with the smallest distance. In each cluster, the expanded points use w to increase randomness.

[0033] Right now .

[0034] Furthermore, in S6, the distance between the cluster centers finally obtained in the K-means clustering process is used to filter the newly added sample points in different classes. This is mainly to prevent the distance between the newly expanded points in different clusters from being too close, that is, the expanded points may have multiple candidate annotation labels because the distance is too close. This step is to prevent too many expanded points, and then the data deviates from the actual usage scenario. The specific method is to take an expanded point in each of the two different clusters, calculate the distance between the two points, and if it is less than the distance between the centers of the two clusters, remove the two points.

[0035] Furthermore, the step SD specifically includes the following steps:

[0036] The driving scenario M1 corresponds to multiple expected functional safety models.

[0037] Furthermore, the step SE specifically includes the following steps:

[0038] If there is no trained model X in the trained models corresponding to the driving scenario M1, the untrained model W is ignored and the process of training a new model is started.

[0039] Furthermore, the step SF specifically includes the following steps:

[0040] Feedback of the trained model X to the Internet of Vehicles platform to allow the platform to confirm whether the model X can meet the requirements is specifically as follows:

[0041] Before the platform server performs model selection or model training, the project manager sets a data volume threshold;

[0042] If the amount of training data of the trained model X reaches the data amount threshold, it is determined to meet the requirement;

[0043] Otherwise, the demand is not met.

[0044] Further, judging whether the amount of training data of the trained model X reaches the data amount threshold specifically comprises the following steps:

[0045] SF1. Obtain the training sample set corresponding to the trained model X;

[0046] SF2. Extracting samples labeled as emergencies and rare driving behaviors from the training sample set;

[0047] Wherein, each training sample in the training sample set corresponds to a label;

[0048] SF3. Count the total number of samples labeled as emergencies and rare driving behaviors in the training sample set, and when the proportion of the total number of samples in the total number of samples in the training sample set is less than a preset threshold When the training data volume of the trained model X does not reach the data volume threshold, it is determined that the demand is not met; when the proportion of the total number of samples in the total number of samples in the training sample set is greater than or equal to the preset threshold and is less than or equal to the preset threshold When the proportion of the total number of samples in ... training sample set is greater than a preset threshold, step SF4 is executed; when the proportion of the total number of samples in the training sample set is greater than a preset threshold When the training data volume of the trained model X reaches the data volume threshold, it is determined that the requirement is met;

[0049] Among them, the preset threshold The preset threshold is 30%. It is 60%.

[0050] SF4. Cluster the samples of emergencies and rare driving behaviors extracted in step SF2 using the K-means algorithm to form multiple cluster sets of emergencies and rare driving behaviors samples { };

[0051] in, is the hth emergency situation and rare driving behavior sample clustering set in multiple emergency situation and rare driving behavior sample clustering sets; u is the total number of clustering sets in multiple emergency situation and rare driving behavior sample clustering sets.

[0052] SF5. Clustering collection for emergencies and rare driving behavior samples For all samples in , calculate the distance between any two sample points in all samples to form multiple sample point distances;

[0053] SF6. Sort the multiple sample point distances formed in step SF5 in ascending order to form a sorting result consisting of sample point distances, and obtain sample pairs corresponding to the top 20% of the sample point distances in the sorting result in ascending order;

[0054] The sorting result is a sequence formed by arranging the sample point distances from small to large;

[0055] The sample pair corresponding to the sample point distance is the two sample points that form the sample point distance;

[0056] SF7. Calculate the center point of each sample pair corresponding to the first 20% of the sample points to form multiple center points;

[0057] The center point of each group of sample pairs is the center point of the two sample points in each group of sample pairs;

[0058] SF8. Calculate the distance between any two center points of the multiple center points in step SF7 respectively to form multiple center point distances, sort the multiple center point distances from small to large, and obtain the minimum value in the sorting result;

[0059] SF9. When the minimum value is greater than the preset threshold When judging When the minimum value is less than or equal to the preset threshold value, When judging The data volume requirement is not met.

[0060] Among them, the preset threshold It can be flexibly set according to the actual scenario.

[0061] SF10. Clustering of multiple emergencies and rare driving behavior samples { }, repeat steps SF5 to SF9. When all u sets of emergency situations and rare driving behavior sample clusters are judged to meet the data volume requirement, it is judged that the training data volume of the trained model X reaches the data volume threshold, and it is judged to meet the requirement; when any one of the u sets of emergency situations and rare driving behavior sample clusters is judged to not meet the data volume requirement, it is judged that the training data volume of the trained model X does not reach the data volume threshold, and it is judged to not meet the requirement.

[0062] Furthermore, if the model X fails to meet the platform requirements, the model X is deleted while starting the process of training a new model.

[0063] The beneficial effects of the present invention are as follows:

[0064] 1. The present invention, through steps S1-S6, expands the expected functional safety test data by combining K-means algorithm clustering, linear interpolation and Gaussian white noise error to expand the expected functional safety test data; the expected functional safety test data expanded by the above method can improve the efficiency of expected functional safety training data collection and generation, and reduce data collection costs. By expanding the data set and increasing the sample proportion of emergencies and rare driving behaviors, the functional safety of autonomous driving is enhanced.

[0065] 2. Through the setting of steps SA-SH, it is possible to determine whether it is necessary to re-collect data and train the model when the Internet of Vehicles platform needs to call or establish a model. If it is not necessary to re-collect data and train the model, the previous perfect model is directly called. If it is necessary to re-collect data and train the model, a less precise method is used for model training. This method can take into account the comprehensiveness of sample data and the effective use of system resources.

[0066] 3. By further refining SF, it is possible to use the amount of training data as a criterion to determine whether further processing is needed, and to accurately and quickly execute the judgment process, saving system resources.

[0067] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above description and other purposes, features and advantages of the present invention more obvious and easy to understand, the preferred embodiments are specifically cited and described in detail as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0069] Figure 1 A flowchart of an expansion in a method for expanding data in an expected functional safety data set is provided. DETAILED DESCRIPTION

[0070] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0071] In the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be connected, detachably connected, or integrated; it can be mechanically connected or electrically connected; it can be directly connected or indirectly connected through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0072] In view of the above problems, the present invention proposes a method for expanding the expected functional safety test data of an Internet of Vehicles, and the method specifically comprises the following steps:

[0073] SA detects a call request of the Internet of Vehicles platform model. If the call request is detected, SB is executed. If the call request is not detected, the Internet of Vehicles platform continues to be monitored;

[0074] SB. If the call request is detected, determining whether the Internet of Vehicles platform is to further improve the trained model or train a new model;

[0075] SC. If a new model is to be trained, the process of training a new model will be started;

[0076] SD. If the trained model is to be further improved, the driving scene M1 corresponding to the model M to be trained is determined;

[0077] SE. Query whether there is a trained model X in the trained models corresponding to the driving scenario M1;

[0078] SF. If the trained model X exists, the trained model X is fed back to the Internet of Vehicles platform to allow the platform to confirm whether the model X can meet the requirements;

[0079] SG. If the model X can meet the platform requirements, directly call the model X and apply it;

[0080] SH. If the model X fails to meet the platform requirements, start the process of training a new model.

[0081] Furthermore, the step SA specifically includes the following steps:

[0082] The Internet of Vehicles platform model calling request is issued by the Internet of Vehicles server.

[0083] Furthermore, the step SB specifically includes the following steps:

[0084] The Internet of Vehicles server can only execute two types of processes: improving the trained model and training a new model.

[0085] Furthermore, the training new model process in step SC is:

[0086] S1. Use K-means algorithm to cluster emergency and rare driving behavior data;

[0087] S2. Perform sample expansion in each of the clusters and execute step S3;

[0088] S3. Use the Euclidean distance to calculate the K similar samples closest to xi;

[0089] S4. Perform linear difference among K similar samples;

[0090] S5. Add Gaussian white noise to each sample generated by interpolation to increase randomness;

[0091] S6. Using the distance of the cluster centers finally obtained in the K-means clustering process, filter the newly added sample points in different classes to form a sample set for completing data expansion;

[0092] S7. Perform model training based on the sample set that has completed data expansion to complete the training of the new model.

[0093] Furthermore, the step SD specifically includes the following steps:

[0094] The driving scenario M1 corresponds to multiple expected functional safety models.

[0095] Furthermore, the step SE specifically includes the following steps:

[0096] If there is no trained model X in the trained models corresponding to the driving scenario M1, the untrained model W is ignored and the process of training a new model is started.

[0097] Furthermore, the step SF specifically includes the following steps:

[0098] Feedback of the trained model X to the Internet of Vehicles platform to allow the platform to confirm whether the model X can meet the requirements is specifically as follows:

[0099] Before the platform server performs model selection or model training, the project manager sets a data volume threshold;

[0100] If the amount of training data of the trained model X reaches the data amount threshold, it is determined to meet the requirement;

[0101] Otherwise, the demand is not met.

[0102] Further, judging whether the amount of training data of the trained model X reaches the data amount threshold specifically comprises the following steps:

[0103] SF1. Obtain the training sample set corresponding to the trained model X;

[0104] SF2. Extracting samples labeled as emergencies and rare driving behaviors from the training sample set;

[0105] Wherein, each training sample in the training sample set corresponds to a label;

[0106] SF3. Count the total number of samples labeled as emergencies and rare driving behaviors in the training sample set, and when the proportion of the total number of samples in the total number of samples in the training sample set is less than a preset threshold When the training data volume of the trained model X does not reach the data volume threshold, it is determined that the demand is not met; when the proportion of the total number of samples in the total number of samples in the training sample set is greater than or equal to the preset threshold and is less than or equal to the preset threshold When the proportion of the total number of samples in ... training sample set is greater than a preset threshold, step SF4 is executed; when the proportion of the total number of samples in the training sample set is greater than a preset threshold When the training data volume of the trained model X reaches the data volume threshold, it is determined that the requirement is met;

[0107] Among them, the preset threshold The preset threshold is 30%. It is 60%.

[0108] SF4. Cluster the samples of emergencies and rare driving behaviors extracted in step SF2 using the K-means algorithm to form multiple cluster sets of emergencies and rare driving behaviors samples { };

[0109] in, is the hth emergency situation and rare driving behavior sample clustering set in multiple emergency situation and rare driving behavior sample clustering sets; u is the total number of clustering sets in multiple emergency situation and rare driving behavior sample clustering sets.

[0110] SF5. Clustering collection for emergencies and rare driving behavior samples For all samples in , calculate the distance between any two sample points in all samples to form multiple sample point distances;

[0111] SF6. Sort the multiple sample point distances formed in step SF5 in ascending order to form a sorting result consisting of sample point distances, and obtain sample pairs corresponding to the top 20% of the sample point distances in the sorting result in ascending order;

[0112] The sorting result is a sequence formed by arranging the sample point distances from small to large;

[0113] The sample pair corresponding to the sample point distance is the two sample points that form the sample point distance;

[0114] SF7. Calculate the center point of each sample pair corresponding to the first 20% of the sample points to form multiple center points;

[0115] The center point of each group of sample pairs is the center point of the two sample points in each group of sample pairs;

[0116] SF8. Calculate the distance between any two center points of the multiple center points in step SF7 respectively to form multiple center point distances, sort the multiple center point distances from small to large, and obtain the minimum value in the sorting result;

[0117] SF9. When the minimum value is greater than the preset threshold When judging When the minimum value is less than or equal to the preset threshold value, When judging The data volume requirement is not met.

[0118] Among them, the preset threshold It can be flexibly set according to the actual scenario.

[0119] SF10. Clustering of multiple emergencies and rare driving behavior samples { }, repeat steps SF5 to SF9. When all u sets of emergency situations and rare driving behavior sample clusters are judged to meet the data volume requirement, it is judged that the training data volume of the trained model X reaches the data volume threshold, and it is judged to meet the requirement; when any one of the u sets of emergency situations and rare driving behavior sample clusters is judged to not meet the data volume requirement, it is judged that the training data volume of the trained model X does not reach the data volume threshold, and it is judged to not meet the requirement.

[0120] Furthermore, if the model X fails to meet the platform requirements, the model X is deleted while starting the process of training a new model.

[0121] The expected functional safety data set is used to train the expected functional safety model so that the vehicle can realize the driving function based on the expected functional safety model.

[0122] The autonomous driving data set includes: one or more of visual camera data, thermal imaging camera data, radar data, lidar data, ultrasonic sensor data, global navigation satellite system data, high-precision map data, and odometer data. And the expected functional safety data set can be divided into target detection data set, semantic segmentation data set, lane detection data set, driving behavior data set, and positioning and map data set.

[0123] The present invention proposes a method for expanding data in the expected functional safety data set in the field of autonomous driving. By expanding the data of emergencies and rare driving behaviors in the original data set, the total amount of data becomes larger.

[0124] In the method for expanding data in the expected functional safety data set, the data of emergencies and rare driving behaviors in the data set are expanded. Compared with normal driving behavior data, the proportion of emergencies and rare driving behavior data in the original data set is small, less than 1%. After data expansion, the increase of emergencies and rare driving behavior data can make the training model more sensitive to emergencies and rare driving behaviors.

[0125] The data defining emergencies and rare driving behaviors is x, and the data of the ith emergencies and rare driving behaviors is xi. The emergencies and rare driving behaviors data used in the extended data must be labeled, that is, the labeled data is used as the basic data. The overall process of extending the data is as follows: Figure 1 As shown:

[0126] S1. Use K-means algorithm to cluster emergency and rare driving behavior data;

[0127] S2. Perform sample expansion in each of the clusters and execute step S3;

[0128] S3. Use the Euclidean distance to calculate the K1 similar samples closest to xi;

[0129] S4. Perform linear interpolation among K1 similar samples;

[0130] S5. Add Gaussian white noise to each sample generated by interpolation to increase randomness;

[0131] S6. Using the distance of the cluster centers finally obtained in the K-means clustering process, filter the newly added sample points in different classes to form a sample set for completing data expansion;

[0132] S7. Perform model training based on the sample set that has completed data expansion to complete the training of the new model.

[0133] Furthermore, in S1, the K-means algorithm is used to cluster the emergency and rare driving behavior data to distinguish how many types of emergency and rare driving behaviors there are in the collected data. Since the basic data has labels, and the newly expanded data must be related to the data in the real world, the data is first classified, and then the newly expanded data can directly carry the labels. In the K-means clustering process, the choice of K determines the convergence speed of the calculation. Therefore, the value of K can be manually defined after manually browsing the data.

[0134] Furthermore, in S2, data are expanded in different clusters respectively. Since the core method of expanding data in this method is linear interpolation and Gaussian white noise error, it is necessary to ensure the consistency and integrity of the expanded data with the actual driving scene. Therefore, the emergency and rare driving behavior data are clustered first, and data expansion is performed in specific areas of data distribution.

[0135] Furthermore, in S3-S4, the Euclidean distance is used to calculate the K1 similar samples that are closest to xi. That is, in each cluster, K1 sample points with the closest distance are selected. Here x i Represents the data of the i-th emergency and rare driving behavior in a cluster. That is, the distance between any two points in the cluster is calculated, and then K1 points with the smallest distance are selected. Interpolation is performed among these K1 points. The core expression of the interpolation process is:

[0136]

[0137] where p ki The k in y is the kth point among the K1 points with the smallest distance, and i is the ith point in a cluster. k Indicates that among all the points in the cluster, the kth point with the smallest distance between any two points is calculated, x i is the point representing the i-th emergency situation and rare driving behavior in a cluster. ki For k With x i The number of z is determined by the scale to be expanded. rand(0,1) represents a random number between 0 and 1.

[0138] Furthermore, in S5, each expanded point is subjected to Gaussian white noise to enhance randomness. The Gaussian white noise expression is as follows:

[0139]

[0140] Where w is Gaussian white noise, that is, the error of Gaussian distribution. μ is the expected value of the distance between the K1 points with the smallest distance, and σ is the standard deviation of the distance between the K1 points with the smallest distance. In each cluster, the expanded points use w to increase randomness.

[0141] Right now .

[0142] Furthermore, in S6, the distance between the cluster centers finally obtained in the K-means clustering process is used to filter the newly added sample points in different classes. This is mainly to prevent the distance between the newly expanded points in different clusters from being too close, that is, the expanded points may have multiple candidate annotation labels because the distance is too close. This step is to prevent too many expanded points, and then the data deviates from the actual usage scenario. The specific method is to take an expanded point in each of the two different clusters, calculate the distance between the two points, and if it is less than the distance between the centers of the two clusters, remove the two points.

[0143] The beneficial effects of the invention are as follows:

[0144] 1. The present invention can provide a method and process for expanding sensor data of expected functional safety training data through steps S1-S6, and expand the expected functional safety data by combining K-means algorithm clustering, linear interpolation and Gaussian white noise error; the expected functional safety data expanded by the above method can improve the efficiency of collecting and generating expected functional safety training data and reduce data collection costs. By expanding the data set and increasing the sample proportion of emergencies and rare driving behaviors, the functional safety of autonomous driving is enhanced.

[0145] 2. Through the setting of steps SA-SH, it is possible to determine whether it is necessary to re-collect data and train the model when the Internet of Vehicles platform needs to call or establish a model. If it is not necessary to re-collect data and train the model, the previous perfect model is directly called. If it is necessary to re-collect data and train the model, a less precise method is used for model training. This method can take into account the comprehensiveness of sample data and the effective use of system resources.

[0146] 3. By further refining SF, it is possible to use the amount of training data as a criterion to determine whether further processing is needed, and to accurately and quickly execute the judgment process, saving system resources.

[0147] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for expanding data of expected functional safety test of Internet of Vehicles, characterized in that: The method specifically comprises the following steps: SA detects a call request of the Internet of Vehicles platform model. If the call request is detected, SB is executed. If the call request is not detected, the Internet of Vehicles platform continues to be monitored; SB. If the call request is detected, determining whether the Internet of Vehicles platform is to further improve the trained model or train a new model; SC. If a new model is to be trained, the process of training a new model will be started; SD. If the trained model is to be further improved, the driving scene M1 corresponding to the model M to be trained is determined; SE. Query whether there is a trained model X in the trained models corresponding to the driving scenario M1; SF. If the trained model X exists, the trained model X is fed back to the Internet of Vehicles platform to allow the platform to confirm whether the model X meets the requirements; SG. If the model X meets the platform requirements, the model X is directly called and applied; SH. If the model X fails to meet the platform requirements, start training a new model; Step SF specifically includes the following steps: Feedback of the trained model X to the Internet of Vehicles platform to allow the platform to confirm whether the model X can meet the requirements is specifically as follows: Before the platform server performs model selection or model training, the project manager sets a data volume threshold; If the amount of training data of the trained model X reaches the data amount threshold, it is determined to meet the requirement; Otherwise, the demand is not met; The step of judging whether the amount of training data of the trained model X reaches the data amount threshold specifically includes the following steps: SF1. Obtain the training sample set corresponding to the trained model X; SF2. Extracting samples labeled as emergencies and rare driving behaviors from the training sample set; Wherein, each training sample in the training sample set corresponds to a label; SF3. Count the total number of samples labeled as emergencies and rare driving behaviors in the training sample set, and when the proportion of the total number of samples in the total number of samples in the training sample set is less than a preset threshold When the training data volume of the trained model X does not reach the data volume threshold, it is determined that the demand is not met; when the proportion of the total number of samples in the total number of samples in the training sample set is greater than or equal to the preset threshold and is less than or equal to the preset threshold When the proportion of the total number of samples in ... training sample set is greater than a preset threshold, step SF4 is executed; when the proportion of the total number of samples in the training sample set is greater than a preset threshold When the training data volume of the trained model X reaches the data volume threshold, it is determined that the requirement is met; Among them, the preset threshold The preset threshold is 30%. 60%; SF4. Cluster the samples of emergencies and rare driving behaviors extracted in step SF2 using the K-means algorithm to form multiple cluster sets of emergencies and rare driving behaviors samples { }; in, is the hth emergency situation and rare driving behavior sample clustering set in multiple emergency situation and rare driving behavior sample clustering sets; u is the total number of clustering sets in multiple emergency situation and rare driving behavior sample clustering sets; SF5. Clustering collection for emergencies and rare driving behavior samples For all samples in , calculate the distance between any two sample points in all samples to form multiple sample point distances; SF6. Sort the multiple sample point distances formed in step SF5 in ascending order to form a sorting result consisting of sample point distances, and obtain sample pairs corresponding to the top 20% of the sample point distances in the sorting result in ascending order; The sorting result is a sequence formed by arranging the sample point distances from small to large; The sample pair corresponding to the sample point distance is the two sample points that form the sample point distance; SF7. Calculate the center point of each sample pair corresponding to the first 20% of the sample points to form multiple center points; The center point of each group of sample pairs is the center point of the two sample points in each group of sample pairs; SF8. Calculate the distance between any two center points of the multiple center points in step SF7 respectively to form multiple center point distances, sort the multiple center point distances from small to large, and obtain the minimum value in the sorting result; SF9. When the minimum value is greater than the preset threshold When judging When the minimum value is less than or equal to the preset threshold value, When judging The data volume requirements are not met; SF10. Clustering of multiple emergencies and rare driving behavior samples { }, repeat steps SF5 to SF9, when all the u emergency situation and rare driving behavior sample cluster sets are judged to meet the data volume requirement, it is judged that the training data volume of the trained model X reaches the data volume threshold, and it is judged that the requirement is met; when any one of the u emergency situation and rare driving behavior sample cluster sets is judged not to meet the data volume requirement, it is judged that the training data volume of the trained model X does not reach the data volume threshold, and it is judged that the requirement is not met; If the model X fails to meet the platform requirements, the model X is deleted while starting the training process of a new model; The process of training a new model in step SC is as follows: S1. Use K-means algorithm to cluster emergency and rare driving behavior data; S2. Perform sample expansion in each of the clusters; S3. Use the Euclidean distance to calculate the K similar samples closest to xi; xi represents the data of the i-th emergency situation and rare driving behavior in a cluster; S4. Perform linear interpolation among K similar samples; S5. Add Gaussian white noise to each sample generated by interpolation to increase randomness; S6. Using the distance of the cluster centers finally obtained in the K-means clustering process, filter the newly added sample points in different classes to form a sample set for completing data expansion; S7. Performing model training according to the sample set for completing data expansion to complete the training of the new model; Wherein, step SE specifically includes the following steps: If there is no trained model X in the trained models corresponding to the driving scenario M1, the untrained model W is ignored and the process of training a new model is started.

2. A method for expanding data of expected functional safety test of Internet of Vehicles according to claim 1, characterized in that: Step SA specifically includes the following steps: The Internet of Vehicles platform model calling request is issued by the Internet of Vehicles server.

3. A method for expanding data of expected functional safety test of Internet of Vehicles according to claim 2, characterized in that: Step SB specifically includes the following steps: The Internet of Vehicles server executes two types of processes: improving the trained model and training a new model.

4. A method for expanding data of expected functional safety test of Internet of Vehicles according to claim 3, characterized in that: Step SD specifically includes the following steps: The driving scenario M1 corresponds to multiple expected functional safety models.

Images