A method for generating typical working conditions of a vehicle based on space-time distribution optimization

Abstract

The application provides a vehicle typical working condition generation method based on space-time distribution optimization, specifically including four steps of database construction, working condition generation, working condition typicality inspection and space-time distribution optimization working condition, historical vehicle driving data or collected data is collected according to self needs of a user, a vehicle working condition database of specific needs is formed, and vehicle driving working conditions are randomly generated in the mode of short trip division, feature extraction, segment combination and the like; the distribution of the generated working conditions is subjected to typicality requirement by using speed and acceleration distribution statistics inspection, space-time distribution optimization based on the Wasserstein distance is combined, so that the generated working conditions can fully represent the features of the database while being capable of flexibly expanding the needs, and finally, a vehicle typical working condition generation method based on space-time distribution optimization is realized, which takes into account universality, randomness and typicality.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle detection, and in particular to a vehicle typical working condition generation method based on space-time distribution optimization. BACKGROUND

[0002] With the development of today's automobile industry at an increasingly fast pace, diversified application scenarios are needed to test various performances of various new technologies when applied to the whole vehicle. At present, units such as vehicle enterprises, research institutes and colleges generally use the following methods to test the effect of new technologies when applied to the whole vehicle in the research and development stage: a) simulation research through various whole vehicle simulation software (such as Amesim, AVL Cruise, etc.); b) experimental research, which includes two types of test methods: 1. using industry-standard working conditions (such as WLTC, CLTC, etc.) to conduct whole vehicle drum bench tests in an environmental chamber; 2. conducting related tests on roads in different regions according to the needs of the vehicle in the research and development stage (such as high-cold, high-temperature and high-altitude tests). The first test method has the disadvantage of single scene, that is, the standard working condition cannot cover most of the actual scenes and cannot meet various research and development needs (the same problem also exists for simulation research); the second test method can meet the research and development needs, but has a long test cycle and high cost. Combining the advantages of the two, using typical working conditions that meet various research and development needs as the input of the simulation environment or conducting bench tests in the laboratory is the common demand of various units, and therefore, inventing a vehicle typical working condition generation method has become the current research focus. In addition, different needs exist when different research methods of simulation and test are used: simulation research has a lower cost, so more attention is paid to the diversity of the working condition input to obtain more regularity conclusions, and therefore, greater randomness is required for the generated working condition; test research pays more attention to the actual performance under the input of a certain scene, and therefore, greater typicality is required for the generated working condition. SUMMARY

[0003] The purpose of the present application is to provide a vehicle typical working condition generation method based on space-time distribution optimization, which takes into account universality, randomness and typicality.

[0004] To achieve the above purpose, the present application provides a vehicle typical working condition generation method based on space-time distribution optimization, comprising the following steps:

[0005] S1: collecting actual running data and RDE test data of the vehicle;

[0006] S2: sequentially performing short-trip segment extraction, short-trip segment feature calculation and classification, and working condition construction based on the data;

[0007] S3: working condition typicality inspection: based on the working condition curves L iis a random result, in order to make the generated working condition can fully represent the characteristics of the database, using the speed- acceleration distribution chi-square test on the distribution of generated working conditions typical requirements;

[0008] S4: working condition space-time distribution optimization based on Wasserstein distance.

[0009] Further, in the S1, the actual running data of the vehicle includes vehicle motion characteristics, power characteristics and environmental characteristics data; collection site, collection road and collection time can be decided according to the user's own needs;

[0010] The collection frequency of the RDE test data is 1-10Hz, and the collection site, collection road and collection time can be decided according to the user's own needs.

[0011] Further, in the S2, the short trip segment extraction: the movement process of the vehicle from the idle state to the next idle start to stop movement is defined as a short trip segment, first from the collected vehicle driving working condition original data to divide and extract short trip segment, at the same time, calculate and summarize the length of each idle segment, form short trip segment database and idle database.

[0012] Further, in the S1, the short trip segment feature calculation is to describe and evaluate the characteristics of each short trip segment by calculating various characteristic values of short trip segment.

[0013] Further, the running time T, the average speed v m , the segment starting slope α start , the segment ending slope α end and the average slope α m are selected to describe and evaluate the characteristics of short trip segment;

[0014] The short trip segment is classified by the interval of the average speed of the short trip segment; after classification, the proportion of the total length of the segment in each category to the total length of the segment in the short trip segment database is calculated i represents the category.

[0015] Further, in the S2,

[0016] The steps of working condition construction are as follows:

[0017] Step 1: according to the user's own needs, the allowed minimum motion time length of the working condition to be constructed is preset as T tar ;

[0018] Step 2: calculate the allowed minimum motion time length of each category in the working condition to be constructed, the calculation method is T tar_i =p i ·Ttar ;

[0019] Step 3: For each category, do the following: a. Initialize the working condition sequence L i of the i-th category with the sequence motion time length t tar_i = 0 and the sequence total time length t i = 0; b. Randomly extract a length t idle_temp and t i_temp of idle and short-trip segments from the idle database and the short-trip segment database of the i-th category respectively, and add them to the end of L i , update t tar_i = t tar_i + t i_temp , t i = t i + t idle_temp + t i_temp ;

[0020] Step 4: If t tar_i < T tar_i , repeat step b and increase the condition that the starting slope of the current segment and the ending slope of the previous segment differ by no more than a limit value a0, a0 can be defined, otherwise terminate the loop and output L i and t i .

[0021] Further, in the S3, the following steps are included:

[0022] For each category, do the following:

[0023] d. Establish the speed-acceleration (v-a) frequency distribution matrix M ref_i of each value in the short-trip segment database of this category;

[0024] e. Construct the working condition curve L i according to the above process;

[0025] f. Establish the v-a frequency distribution matrix M i of the curve L i , and establish the same requirements as M ref_i ;

[0026] d. Take M ref_i as the theoretical distribution and M i as the observed frequency distribution, and perform chi-square goodness-of-fit test; e. If the significance level p of the chi-square test result is less than 0.05, repeat steps b, c, d, otherwise terminate the loop and output L i .

[0027] Furthermore, in S4, the Wasserstein distance measures the minimum average distance that needs to be moved when shifting data from distribution p to distribution q, and can quantitatively represent the similarity between two distributions; the formal expression of the Wasserstein distance is as follows:

[0028]

[0029] Here, Π(p,q) represents the set of all possible joint distributions combining distributions p and q; for each possible joint distribution γ, a sample x and y can be obtained by sampling (x,y) ~ γ, and the distance ||xy|| between these samples can be calculated. Therefore, the expected value E of the distance between the sample pairs under this joint distribution γ can be calculated. x,y～γ [||xy||]; the lower bound that this expectation value can be obtained in all possible joint distributions is the Wasserstein distance.

[0030] Furthermore, in S4, the specific steps for optimizing the spatiotemporal distribution of the working conditions based on the Wasserstein distance are as follows:

[0031] Step 1: Referring to the frequency distribution calculation method, determine the intervals for the speed interval, acceleration interval, and gradient interval respectively. It is recommended to use intervals of 1 km / h and 0.1 m / s². 2 The velocity distribution, acceleration distribution, and slope distribution are constructed using intervals of 0.001, which can also be defined according to the actual database characteristics and requirements.

[0032] Step 2: For the operating condition curve L, calculate the velocity frequency distribution p of all motion segments in the curve. v Acceleration frequency distribution p a and slope frequency distribution p α Simultaneously calculate the velocity frequency distribution q of the short-stroke segment database. v Acceleration frequency distribution q a and slope frequency distribution q α ;

[0033] Step 3: Calculate W(p) v ,q v ), W(p a ,q a ) and W(p α ,q α Custom W(p) v ,q v ), W(p a ,q a ) and W(p α ,q α The limit W v Wa and W α ;

[0034] Step 4, if W(p v , q v )≤W v and W(p a , q a )≤W a and W(p α , q α )≤W α are not met, a new working condition curve L is constructed according to the flow again and steps 2 and 3 are repeated, otherwise the cycle is terminated and the working condition curve L is output.

[0035] Compared with the prior art, the advantages of the present application are that:

[0036] 1. The method of the present application is not specific to a particular database, and users can construct it according to their own needs according to the database construction steps of the present application, which has good versatility.

[0037] 2. After generating the working condition curve, the present application uses speed and acceleration distribution statistical tests to require typicality of the generated working condition distribution, so that the generated working condition can fully represent the characteristics of the database.

[0038] 3. The spatiotemporal distribution optimization based on Wasserstein distance of the present application provides multiple adjustable parameters, so that users can adjust the randomness and typicality requirements of the generated results according to their needs, which has good flexibility.

[0039] 4. Through the working condition generation method of the present application, users can promote the change of the research and development process from actual road to whole vehicle hub test according to market demand and national requirements, and provide a benchmark for whole vehicle simulation development, thereby reducing research and development cost and shortening research and development cycle. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 is a short trip segment classification result in the embodiment of the present application.

[0041] Figure 2 is an L2 construction result in the embodiment of the present application.

[0042] Figure 3 is the speed frequency distribution of a working condition curve in the embodiment of the present application.

[0043] Figure 4 is the automobile typical working condition generation result in the embodiment of the present application. DETAILED DESCRIPTION

[0044] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions of the present application will be further described below.

[0045] The present application designs a typical vehicle operating condition generation method based on space-time distribution optimization. A user collects historical vehicle driving data or acquires data according to his own needs to form a vehicle operating condition database with specific needs, and randomly generates vehicle driving operating conditions by using short trip division, feature extraction, segment combination and other methods. The distribution of the generated operating conditions is subjected to statistical test of speed and acceleration distribution to meet the typicality requirement, and combined with space-time distribution optimization based on Wasserstein distance, so that the generated operating conditions can fully represent the characteristics of the database while being flexible to expand the needs, and finally realize a typical vehicle operating condition generation method based on space-time distribution optimization which takes into account the generality, randomness and typicality.

[0046] Specifically, the following steps are included:

[0047] 1. Data acquisition

[0048] The user can collect data according to his own needs. The data sources include but are not limited to the following two types:

[0049] 1) Vehicle actual operation data

[0050] The vehicle motion characteristics, power characteristics and environmental characteristics data of the vehicle during actual operation are collected. The collection location can be in one or each region / city according to the needs; the collection road includes but is not limited to urban area, suburban area, trunk road, branch road, expressway, secondary trunk road and highway, etc.; the collection time includes but is not limited to spring, summer, autumn and winter, weekdays, holidays, peak hours and off-peak hours.

[0051] 2) RDE test data

[0052] The RDE test data of the vehicle is collected, and the collection location, collection road and collection time are the same as above.

[0053] The guiding suggestions for data collection include: the sampling frequency of data is recommended to be 1-10 Hz, and the total sampling time, collection location, collection road and collection time can be determined according to the user's own needs, but in order to ensure the typicality and randomness of the generated operating conditions, the total sampling time should not be too short, and the collection location, collection road and collection time type should not be too single.

[0054] In this embodiment, the data of a certain type of passenger car in a certain city during actual operation is collected, with a total mileage of about 11100km, a sampling frequency of 1Hz, and collected data including driving time, speed, acceleration, etc., and environmental characteristics including road mileage, slope, etc. (It is intended to indicate that if other people collect additional data, but the subsequent working condition generation process is similar to this invention, it is also considered within the protection scope of this patent).

[0055] 2. Working condition generation process

[0056] 1) Short trip segment extraction

[0057] This invention refers to existing literature and defines the movement process of a vehicle from idle state to the next idle start stop movement as a short trip segment. First, the short trip segments are divided and extracted from the collected vehicle driving working condition original data, and the duration of each idle segment is calculated and summarized to form a short trip segment database and an idle database.

[0058] In order to facilitate the rationality of subsequent generation of typical working conditions, data cleaning is required for the extracted short trip segments. In this embodiment, invalid short trip segments are defined according to the following conditions, and short trip segments meeting the following conditions are deleted from the database. When using this invention, data cleaning can also be performed according to self-defined requirements.

[0059] 1. Running time greater than 30 minutes;

[0060] 2. Running time less than 5 seconds;

[0061] 3. Maximum acceleration greater than 4.5m / s 2 ;

[0062] 4. Maximum speed less than 5km / h.

[0063] After the above conditions are screened, a total of 7163 short trip segments are obtained in this embodiment.

[0064] 2) Short trip segment feature calculation and classification

[0065] The characteristics of each short trip segment are described and evaluated by calculating various characteristic values of the short trip segment. In this embodiment, running time T(s), average speed v m (km / h), segment starting slope α start (dimensionless), segment ending slope α end (dimensionless), and average slope α m (dimensionless) are selected to describe and evaluate the characteristics of the short trip segment. When using this invention, the definition and calculation of features can also be performed according to self-defined requirements.

[0066] The short-trip segments in this embodiment are classified by the interval of the average speed of the short-trip segments, and the classification rules are shown in the following table.

[0067] Table 1 Classification rules of short-trip segments

[0068]

[0069] In actual use of the present application, the short-trip segments can also be classified based on features according to the features of the collected data and self-defined requirements, and the classification methods include but are not limited to principal component analysis, cluster analysis and other data mining methods.

[0070] After the short-trip segments in the short-trip segment database of this embodiment are classified according to the rules in Table 1, the short-trip segments in each category are as shown in the figure description. Figure 1 The proportion of the total length of the segments in each category to the total length of the segments in the short-trip segment database after the classification is completed is calculated. i represents the category. In this embodiment, i = 1, 2, 3, 4, 5, 6.

[0071] 3) Working condition construction algorithm

[0072] The steps of working condition construction are as follows:

[0073] 1. The user presets the allowed minimum motion time length (i.e. excluding idle speed segments) of the working condition to be constructed as T tar , according to the user's own needs. In this embodiment, T tar is set to 1500s.

[0074] 2. The allowed minimum motion time length of each category in the working condition to be constructed is calculated, and the calculation method is T tar_i = p i · T tar . In this embodiment, i = 1, 2, 3, 4, 5, 6. Taking category 2 as an example, the calculation result of T tar_2 in this embodiment is 394s.

[0075] 3. For each category, the following operations are performed: a. initialize the working condition sequence L i of the i-th category, the sequence motion time length t tar_i = 0 and the sequence total time length t i = 0; b. respectively randomly extract an idle speed segment and a short-trip segment of the i-th category from the idle speed database and the short-trip segment database of the i-th category, with lengths of t idle_temp and t i_temp , and sequentially add them to the end of L i , update t tar_i = t tar_i + t i_temp , t i= t i + t idle_temp + t i_temp ; c. If t tar_i < T tar_i , repeat step b and increase the condition (i.e. the difference between the current segment start slope and the last segment end slope is not more than a limit value a0, a0 can be defined by user, in this embodiment, a0 is recommended to be 0.01), otherwise terminate the loop and output L i and t i . For this embodiment, taking category 2 as an example, a0 is set to be 0.01, and a piece of curve L2 constructed according to this algorithm is shown in FIG. 2. Figure 2

[0076] 3. Typicality inspection of working conditions

[0077] According to the above process, the working condition curves L i of each category are constructed, which are random results. In order to make the generated working conditions fully represent the characteristics of the database, a chi-square test of speed-acceleration distribution is used to require the distribution of the generated working conditions to be typical, and the specific steps are as follows:

[0078] 1. For each category, the following operations are performed: a. Establish a speed-acceleration (v-a) frequency distribution matrix M ref_i for each value in the short-travel segment database under this category, specifically, the speed interval is recommended to be 5 km / h, and the acceleration interval is recommended to be 0.5 m / s 2 ; b. Construct the working condition curve L i according to the above process; c. Establish the v-a frequency distribution matrix M i of the curve L i , and the establishment requirement is the same as M ref_i ; d. Take M ref_i as the theoretical distribution and M i as the observed frequency distribution to perform chi-square goodness-of-fit test; e. If the significance level p of the chi-square test result is less than 0.05, repeat steps b, c, and d, otherwise terminate the loop and output L i .

[0079] Specifically, taking category 2 in this embodiment as an example, the sampling frequency is 1 Hz, and M ref_i can be represented according to the recommended interval as follows:

[0080]

[0081]

[0082] After the working condition curve L i is constructed according to step b, L​i v-a frequency distribution matrix M i , calculate the chi-square test statistics (χ 2 ) between the two matrices, according to χ 2 value and degree of freedom to query the significance level value of chi-square distribution. In this embodiment, the degree of freedom is (4-1) x (6-1) = 15.

[0083] 2, arrange all the L i s that meet the above working condition typicality test in a random order to form the final working condition curve L, and the time length T of the working condition is T = ∑t i .

[0084] 4, working condition space-time distribution optimization based on Wasserstein distance

[0085] The working condition curve L established by the above process can meet the space-time distribution requirements of the database in statistics, but when the database is large, more different working condition curves can still be generated. On the basis of statistical testing, the application adds a quantitative index of space-time distribution based on Wasserstein distance, and the user can set the parameter value of the index according to the demand to customize the randomness and typicality bias of the result.

[0086] Wasserstein distance measures the minimum average distance required to move data from distribution p to distribution q, so it can quantitatively represent the similarity of two distributions. The formal expression of Wasserstein distance is as follows:

[0087]

[0088] Where Π(p, q) represents the set of all possible joint distributions of distributions p and q combined. For each possible joint distribution γ, a sample x and y can be obtained from x, y ~ γ, and the distance ||x-y|| of the sample pair can be calculated, so the expected value E x,y～γ [||x-y||] of the sample pair distance under the joint distribution γ can be calculated. The lower bound of this expected value that can be taken in all possible joint distributions is the Wasserstein distance.

[0089] The specific steps of working condition space-time distribution optimization based on Wasserstein distance are as follows:

[0090] 1. Refer to the frequency distribution calculation method in the "working condition typicality test" part of the application, determine the interval of the speed interval, acceleration interval and slope interval respectively, and recommend 1 km / h, 0.1 m / s 2The subsequent velocity distribution, acceleration distribution and slope distribution are constructed with intervals of 0.001, and can also be defined according to actual database characteristics and requirements.

[0091] 2. For the working condition curve L constructed according to the above process of the present application, the velocity frequency distribution p v , the acceleration frequency distribution p a and the slope frequency distribution p α of all motion segments in the curve are calculated; meanwhile, the velocity frequency distribution q v , the acceleration frequency distribution q a and the slope frequency distribution q α of the short stroke segment database are calculated.

[0092] 3. W(p v , q v ), W(p a , q a ) and W(p α , q α ) are calculated, and the limits W v , W a and W α of W(p v , q v ), W(p a , q a ) and W(p α , q α ) are defined.

[0093] 4. If W(p v , q v )≤W v , W(p a , q a )≤W a and W(p α , q α )≤W α are not satisfied, a new working condition curve L is constructed according to the process of the present application, and steps 2 and 3 are repeated, otherwise the cycle is terminated and the working condition curve L is output.

[0094] In this embodiment, W v , W a and W α are taken as 1, +∞ and +∞ respectively, that is, only the velocity distribution is optimized, and the acceleration and slope distribution are not optimized. In the optimization process, the velocity distribution is constructed with intervals of 3.6 km / h, and the velocity frequency distribution p v of a certain working condition curve L is shown in Figure 3 of the figure description. After the working condition space-time distribution optimization process based on Wasserstein distance, the finally generated typical working condition curve of the automobile is shown inFigure 4 as shown.

[0095] Prospect of the application

[0096] The application aims at the diversified requirements of input and application scene in the development of automobile field and the insufficient and limitation of the current automobile working condition generation method in the combination of universality, randomness and typicality, and designs a kind of automobile typical working condition generation method based on space-time distribution optimization. The method obtains automobile typical working conditions meeting the statistical space-time homodistribution requirements through data acquisition, segment extraction, feature calculation and classification, working condition construction and other steps combined with the chi-square test of speed-acceleration distribution. Moreover, in the actual use of the method to generate automobile typical working conditions, the working condition space-time distribution optimization module based on Wasserstein distance is introduced, the Wassertein distance of the generated working conditions and the user data set is calculated in the self-defined dimension, the working condition set meeting the current requirements is identified by setting the working condition space-time distribution optimization conditions and parameters, so as to maximize the satisfaction of the user's randomness and typicality demand bias for the generation result.

[0097] The above is only the preferred embodiment of the application, and does not limit the application in any way. Any person skilled in the art can make any form of equivalent replacement, modification or change to the technical solutions and technical contents disclosed by the application without departing from the scope of the technical solutions of the application, which still belongs to the protection scope of the application.

Claims

1. A method for generating typical working conditions of a vehicle based on optimization of space-time distribution, characterized in that, The method comprises the following steps: S1: collecting actual running data and RDE test data of the vehicle; S2: sequentially performing short-trip segment extraction, short-trip segment feature calculation and classification, and working condition construction based on the data; S3: Working condition typicality verification: based on the working condition curves of each category built in S2 is a random result, in order to make the generated working condition fully represent the characteristics of the database, the speed-acceleration distribution chi-square test is used to require the distribution of the generated working condition to be typical; S4: working condition space-time distribution optimization based on Wasserstein distance; In the S2, the short-trip segment feature calculation is to describe and evaluate the characteristics of each short-trip segment by calculating various characteristic values of the short-trip segment; select run time average speed segment start grade segment end grade and average grade to describe and evaluate the characteristics of short trip segments; classifying the short-trip segments by intervals in which the average speeds of the short-trip segments lie; and calculating, after the classifying, a proportion of a total time length of the segments in each category to a total time length of the segments in the short-trip segment database , i represents a category; In the S4, the working condition space-time distribution optimization based on Wasserstein distance comprises the following specific steps: Step 1: Determine the interval of speed interval, acceleration interval and slope interval respectively according to the frequency distribution calculation method, and construct the subsequent speed distribution, acceleration distribution and slope distribution according to the interval of 1 km / h, 0.1 m / s 2 and 0.001, or define it yourself according to the actual database characteristics and needs; Step 2: For the work profile Calculate the speed frequency distribution , the acceleration frequency distribution and the slope frequency distribution for all the moving segments in the profile; simultaneously calculate the speed frequency distribution , the acceleration frequency distribution and the slope frequency distribution for the short-travel segment database; Step 3, calculation , and , customizing limits of , and , and ;​ Step 4, if not satisfied and and ", re-construct the new operating curve according to the flow and repeat steps 2 and 3, otherwise terminate the loop and output the operating curve .

2. The method for generating typical working conditions of a vehicle based on optimization of space-time distribution according to claim 1, characterized in that, In the S1, the actual running data of the vehicle comprises vehicle motion characteristics, power characteristics, and environmental characteristic data; the collection location, collection road, and collection time can be determined according to the user's own needs; The collection frequency of the RDE test data is 1-10 Hz, and the collection location, collection road, and collection time can be determined according to the user's own needs.

3. The method for generating typical working conditions of a vehicle based on optimization of space-time distribution according to claim 1, characterized in that, In the S2, the short-trip segment extraction: the motion process of the vehicle from the idle state to the next idle start to stop motion is defined as a short-trip segment, and first, the short-trip segment is divided and extracted from the collected vehicle driving condition original data, and the time length of each idle segment is calculated and summarized to form a short-trip segment database and an idle database.

4. The method for generating typical working conditions of a vehicle based on optimization of space-time distribution according to claim 1, characterized in that, In the S2, The steps of the working condition construction are as follows: Step 1: the user presets the allowed minimum motion time length of the to-be-constructed working condition according to the user's own needs ; Step 2: Calculate the allowed minimum motion time length of each category in the to-be-constructed working condition, and the calculation method is ; Step 3: For each class, do the following: a. Initialize the sequence of operating conditions for the ith class , the length of the sequence of motion , and the total length of the sequence ; b. Randomly extract a segment of length from the idling database and the short-stroke segment database of the i-th class, respectively. and Add the idling and short-stroke segments to the appropriate locations in sequence. At the end, update , ; Step 4: If , repeat step b and add the condition that the current segment start slope differs from the last segment end slope by no more than a limit , which can be user-defined, otherwise terminate the loop and output and .

5. The method for generating typical working conditions of a vehicle based on optimization of space-time distribution according to claim 4, characterized in that, In the S3, the following steps are included: For each category, the following operations are performed: establishing a speed-acceleration (v-a) frequency distribution matrix for each value in the database of short-range segments under the category ; Constructing the operating curve according to the above procedure ; establishing a curve v-a frequency distribution matrix , establishing requires and the same; d. If the significance level of the chi-squared test is less than a predetermined threshold value, repeat steps b, c, d, otherwise terminate the loop and output as the theoretical distribution, as the observed frequency distribution, perform a chi-squared goodness of fit test;e. If the significance level of the chi-squared test is less than a predetermined threshold value, repeat steps b, c, d, otherwise terminate the loop and output as the theoretical distribution. as the observed frequency distribution.

6. The method for generating typical working conditions of a vehicle based on optimization of space-time distribution according to claim 5, characterized in that, In the S4, the Wasserstein distance measures the minimum value of the average distance required to move the data from distribution p to distribution q, and can quantitatively represent the similarity of two distributions; the formal expression of the Wasserstein distance is as follows: ； where denotes the set of all possible joint distributions p and q can be combined; for each possible joint distribution γ, one can sample a sample x and y, and compute the distance of this pair of samples So one can compute the expected value of the distance of a pair of samples under the joint distribution γ The lower bound that can be taken on this expected value over all possible joint distributions is the Wasserstein distance.

Images