A driving cycle development method and system for large sample size

Abstract

The application discloses a kind of driving condition development method and system for large sample quantity, mainly related to the field of transportation and planning.It includes the following steps: collecting real vehicle driving data and dividing into short trip segment samples and counting its driving characteristics;Analysis and processing of the driving characteristics of short trip segment samples;Output random condition.The beneficial effects of the present application are that it can be applied to classify large sample data, and can construct typical driving conditions for fine classification features.

Description

Technical Field

[0001] This invention relates to the field of transportation and planning, specifically to a development method and system for large-sample driving conditions. Background Technology

[0002] A typical driving condition is a driving curve composed of time-speed data, reflecting the typical operating characteristics of a particular vehicle or group of vehicles under specific traffic conditions. Constructing typical driving conditions enables the goal of reflecting the vehicle operating patterns behind a large amount of data with a small amount of data. It can be used in research and development or testing scenarios such as vehicle dynamics simulation and static emission testing of whole vehicles, and is an important part of assisting vehicle technology development.

[0003] Due to the diverse uses of vehicles and the varied traffic operation scenarios, it is difficult to describe all traffic objects with a single typical operating condition. More targeted development of typical driving conditions is needed to accurately reflect the driving characteristics of specific sub-sectors. In practice, massive amounts of data are often collected to comprehensively cover vehicle usage characteristics. However, with large sample sizes and numerous categories, a limited number of clusters can easily mask more granular features, and overlap between categories is likely. Therefore, it is necessary to delve deeper into and mine the sub-category features inherent in large sample spaces to provide more accurate input for modeling and simulating vehicle driving characteristics. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for developing driving conditions for large sample sizes. It is applicable to classifying large sample data and can construct typical driving conditions based on the characteristics of subdivided categories.

[0005] To achieve the above objectives, the present invention employs the following technical solution:

[0006] A method for developing driving conditions for large sample sizes includes the following steps: collecting real vehicle driving data and segmenting it into short-stroke segment samples and statistically analyzing its driving characteristics; analyzing and processing the driving characteristics of the short-stroke segment samples; and outputting random driving conditions.

[0007] Preferably, the short-stroke segment sample consists of an idle speed segment followed by a driving segment with a final speed of 0, including the following four states: acceleration state: a driving state with a non-zero speed and an acceleration greater than or equal to 0.15 m / s2; deceleration state: a driving state with a non-zero speed and an acceleration less than or equal to -0.15 m / s2; constant speed state: a driving state with a non-zero speed and an acceleration within the range of [-0.15, 0.15] m / s2; and idle speed state: a driving state in which the engine is continuously running but both the acceleration and the vehicle speed are 0.

[0008] Preferably, the driving characteristics of the statistically collected short-distance segment samples are specifically as follows: statistically analyzing the driving characteristics of each short-distance segment sample and constructing a short-distance segment sample feature matrix includes the following steps: statistically analyzing the characteristics of each short-distance sample, the statistical characteristics including: idling time, driving time, acceleration time, deceleration time, constant speed time, maximum speed, average speed of the driving segment, standard deviation of speed of the driving segment, standard deviation of acceleration of the driving segment, maximum acceleration, average acceleration of the acceleration segment, minimum deceleration, average deceleration of the deceleration segment, acceleration ratio of the driving segment, deceleration ratio of the driving segment, and constant speed ratio of the driving segment; filtering short-distance samples, and the actual driving time of each sample after removing the idling time is greater than or equal to 20 seconds; constructing the feature matrix of the short-distance samples, i.e., the short-distance sample dataset, specifically: each row represents a sample, and each column represents a feature parameter.

[0009] Preferably, the analysis and processing of the driving characteristics of short-distance segment samples specifically includes the following steps: selecting three feature parameters—acceleration time, deceleration time, and constant speed time—for principal component analysis; determining a suitable driving time range so that the corresponding principal component sample set conforms to a normal distribution; filtering short-distance segment samples according to the determined driving time range, performing principal component analysis again to obtain the principal component sample dataset, and using the mean shift algorithm to cluster the principal component samples to obtain cluster centers and convert the coordinates of the cluster centers into expected class feature parameter values; dividing the state according to driving speed, counting the number of samples in different state transitions, and using the improved Kneser-Ney algorithm to estimate the state transition probability matrix.

[0010] Preferably, the step of selecting three feature parameters—acceleration time, deceleration time, and constant speed time—for principal component analysis specifically involves: selecting sample data of the three feature parameters—acceleration time, deceleration time, and constant speed time—from the short-stroke segment sample feature matrix for principal component operation, and filtering the number of principal components. The filtering rule is that the cumulative contribution rate of the principal components is not less than 70%, and finally obtaining the filtered principal component sample set.

[0011] Preferably, determining a suitable driving time range so that the corresponding principal component sample set conforms to a normal distribution specifically involves: selecting a driving time range with the duration increasing from short to long, with an initial span greater than or equal to 10 seconds; selecting sample data from the selected principal component sample set based on the time range; randomly selecting an appropriate number of samples from the selected samples, with the sample number accounting for about 30% of the total, and plotting a gamma diagram accordingly. When the scatter distribution of the sample points is close to a straight line with a slope of 1, the selected time range is accepted; otherwise, the time range span is increased by 5 seconds, and sample data is selected from the selected principal component sample set again until the target driving time range is obtained.

[0012] Preferably, the step of filtering short-distance segment samples according to the determined travel time range, performing principal component analysis again to obtain the principal component sample dataset, and using the mean shift algorithm to cluster the principal component samples to obtain cluster centers and convert the coordinates of the cluster centers into expected class feature parameter values ​​specifically involves selecting 11 feature parameters—acceleration time, deceleration time, constant speed time, maximum speed, average speed of the travel segment, standard deviation of speed of the travel segment, standard deviation of acceleration of the travel segment, maximum acceleration, average acceleration of the acceleration segment, minimum deceleration, and average deceleration of the deceleration segment—to participate in the principal component operation. Before the principal component operation, each sample is centered according to the following formula:

[0013]

[0014] Where: μ is the statistical mean of the sample set, σ is the statistical standard deviation of the sample set, x is the feature value of the short-range segment sample, z is the feature value of the short-range segment sample after alignment, k represents the sample number, and i represents the feature parameter number;

[0015] Randomly select several samples from the principal component samples as seed points; iteratively calculate the convergence position using the following formula:

[0016]

[0017] Where: S represents the principal component sample dataset, s represents the sample vector in the principal component sample dataset, T represents the seed point dataset, and t represents the seed point coordinate vector. Let m(T) represent the updated seed point position coordinate vector, and m(T) represent the set of all updated seed point position coordinate vectors. The kernel function K is defined as: The weighting coefficient w is defined as:

[0018] The distance norm of the seed point coordinates for the two iterations is calculated using the following formula:

[0019]

[0020] Where: k represents the coordinate dimension, d represents the value under the specific coordinate dimension, old and new represent the previous iteration and the current iteration, respectively, and N represents the total number of dimensions;

[0021] The loop exits when the distance norm is less than the specified deviation, outputting the latest seed point convergence coordinates; otherwise, the seed point dataset is updated (m(T)→T), and the loop returns to the previous iteration to calculate the convergence position and iterate again. The seed point convergence coordinates are then converted into the desired feature parameter values ​​as class features. The conversion formula is as follows:

[0022]

[0023] Where e1, e2, ..., e p Let c represent the coefficient vector of each principal component sample, where μ and σ represent the statistical mean and standard deviation of the principal component sample dataset, respectively. 1k c 2k c pk This represents the convergence coordinates of the k-th seed point, where p represents the number of feature dimensions. This represents the expected feature of the k-th seed point.

[0024] Preferably, the step of dividing the states according to driving speed, counting the number of samples for different state transitions, and estimating the state transition probability matrix using the improved Kneser-Ney algorithm specifically involves converting vehicle speed into driving states, using the following conversion formula:

[0025] st = ceil(v / DS) + 1

[0026] Where v represents vehicle speed, DS represents the vehicle speed range corresponding to a unit state, the operator ceil(·) represents rounding up, and st represents the state value corresponding to the current vehicle speed;

[0027] The improved Kneser-Ney smoothing algorithm is used to estimate the state transition probability matrix. The estimation formula is as follows:

[0028]

[0029] Where i and j represent the current and next driving states, respectively, c(ij) represents the number of samples corresponding to the ordered binary state combination ij between two adjacent time points, the denominator on the right side of the equals sign counts the number of samples of state i at the current time, D<·> represents the discount factor, γ(i) represents the interpolation coefficient related to state i, and p(j) represents the probability of unary state j. This represents the state transition probability of the ordered binary state combination ij estimated by the algorithm;

[0030] The final estimated state transition probability matrix model is as follows:

[0031]

[0032] In this context, the first index represents the current state value, the second index represents the next state value, and k is the number of states.

[0033] Preferably, the output random operating condition specifically comprises: generating random operating conditions that meet the accuracy requirements using the Markov chain principle, including the following steps: using the average idling time of statistical samples as the expected idling time; calculating the sum of acceleration time, deceleration time, and constant speed time in the obtained expected feature values ​​to obtain the expected non-idling time; adding the expected idling time and the expected non-idling time to obtain the total expected operating condition duration; randomly generating driving conditions using the state transition probability matrix; and calculating the average relative deviation between the driving conditions and the expected features of the feature parameters, using the following formula:

[0034]

[0035] Where V represents the total number of feature parameters, the selected feature parameters include maximum speed, average speed of the driving segment, standard deviation of speed of the driving segment, standard deviation of acceleration of the driving segment, maximum acceleration, average acceleration of the acceleration segment, minimum deceleration, average deceleration of the deceleration segment, acceleration ratio of the driving segment, deceleration ratio of the driving segment, and constant speed ratio of the driving segment. Among them, the last three time ratio parameters are calculated with corresponding acceleration times. Z represents the statistical mean of the k-th feature parameter of the current class. k This represents the expected mean of the k-th feature parameter of the current component sample;

[0036] Calculate the relative percentage deviation between the total driving time and non-idling time and the expected total time and expected non-idling time; output the driving condition when the average relative deviation of the characteristic parameters and the relative deviation of the duration are both less than 5%; otherwise, return to the above steps to iterate again.

[0037] A development system for driving conditions with a large sample size includes a data acquisition unit, a data processing unit, and a data output unit. The data acquisition unit is data-connected to the data processing unit, and the data processing unit is data-connected to the data output unit. The data acquisition unit is used to collect real vehicle driving data, segment it into short-distance segment samples, and statistically analyze its driving characteristics. The data processing unit is used to analyze and process the driving characteristics of the short-distance segment samples. The data output unit is used to output random driving conditions.

[0038] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0039] In the clustering process, the sample feature space is first subdivided using a combination of parameters including driving time, acceleration time, deceleration time, and constant speed time, and the rationality of the classification is evaluated using a normal distribution model. Then, principal component analysis and the mean shift algorithm are used to find the centroids of the subdivided categories. This algorithm has a fast convergence speed, and its convergence is independent of the initial seed point position, exhibiting good stability. Furthermore, all principal component dimensions are included in the clustering operation, preserving the integrity of the sample information. A Markov chain method is used to construct typical driving conditions, where an improved Kneser-Ney algorithm is used to estimate the state transition probability matrix, which can fully account for low-probability events and has broad coverage of different driving conditions. In summary, this method can effectively address the spatial distribution ambiguity of large sample data, and the clustering results have relatively clear interpretability, making it suitable for driving condition development under big data conditions. Attached Figure Description

[0040] Figure 1 This is a flowchart of the method of the present invention.

[0041] Figure 2 This is a scatter plot of the samples and their randomly selected samples within a specified driving time range according to the present invention.

[0042] Figure 3 This is the gamma distribution diagram of the screening samples according to the present invention.

[0043] Figure 4 This is a scatter plot of the sample and seed points before clustering, as presented in this invention.

[0044] Figure 5 This is a scatter plot of the samples and convergence points after clustering, based on the present invention.

[0045] Figure 6 This is the histogram of the state transition probability matrix of this invention.

[0046] Figure 7 This is a construction condition curve diagram of the present invention. Detailed Implementation

[0047] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined in this application.

[0048] Figure 1 The complete process for constructing a method for typical driving conditions includes the following steps:

[0049] S1: Collect real vehicle driving data, segment it into short-stroke segment samples, and statistically analyze its driving characteristics;

[0050] S2: Analyze and process the driving characteristics of short-distance segment samples;

[0051] S3: Output random operating conditions.

[0052] The short-stroke segment sample consists of an idle speed segment followed by a driving segment with a final speed of 0, including the following four states:

[0053] Acceleration state: A driving state with a non-zero speed and an acceleration greater than or equal to 0.15 m / s²;

[0054] Deceleration state: A driving state with a non-zero speed and an acceleration less than or equal to -0.15 m / s²;

[0055] Uniform speed state: A driving state with a non-zero velocity and acceleration within the range of [-0.15, 0.15] m / s²;

[0056] Idle state: The driving state in which the engine is running continuously but the acceleration and vehicle speed are both 0.

[0057] The specific steps for statistically collecting the driving characteristics of short-distance travel segment samples are as follows: statistically analyzing the driving characteristics of each short-distance travel segment sample and constructing a short-distance travel segment sample feature matrix include the following steps:

[0058] S11: Statistically analyze the characteristics of each short-stroke sample, including: idling time, driving time, acceleration time, deceleration time, constant speed time, maximum speed, average speed of driving segment, standard deviation of speed of driving segment, standard deviation of acceleration of driving segment, maximum acceleration, average acceleration of acceleration segment, minimum deceleration, average deceleration of deceleration segment, acceleration ratio of driving segment, deceleration ratio of driving segment, and constant speed ratio of driving segment.

[0059] S12: Select short-trip samples, and the actual driving time of each sample after removing the idling time is greater than or equal to 20 seconds;

[0060] S13: Construct the feature matrix of short-run samples, i.e., the short-run sample dataset, specifically: each row represents a sample, and each column represents a feature parameter.

[0061] The analysis and processing of the driving characteristics of short-distance segment samples specifically includes the following steps:

[0062] S21: Select three characteristic parameters—acceleration time, deceleration time, and constant speed time—for principal component analysis;

[0063] S22: Determine a suitable range of travel times so that the corresponding principal component sample set conforms to a normal distribution;

[0064] S23: Filter short-trip segment samples according to the determined travel time range, perform principal component analysis again to obtain principal component sample dataset, and use the mean shift algorithm to cluster the principal component samples to obtain cluster center points and convert the coordinates of the cluster center points into expected class feature parameter values;

[0065] S24: Divide the states according to driving speed, count the number of samples for different state transitions, and use the improved Kneser-Ney algorithm to estimate the state transition probability matrix.

[0066] The specific steps of step S21 are as follows: Select sample data with three feature parameters (acceleration time, deceleration time, and constant speed time) from the short-stroke segment sample feature matrix, perform principal component operation, and filter the number of principal components. The filtering rule is that the cumulative contribution rate of the principal components is not less than 70%, and finally obtain the filtered principal component sample set.

[0067] Step S22 specifically involves: selecting a travel time range, with the duration increasing from short to long, and the initial span being greater than or equal to 10 seconds; selecting sample data from the selected principal component sample set based on the time range; randomly selecting an appropriate number of samples from the selected samples, with the sample number accounting for about 30% of the total, and plotting a gamma diagram accordingly. When the scatter distribution of the sample points is close to a straight line with a slope of 1, the selected time range is accepted; otherwise, the time range span is increased by 5 seconds, and sample data is selected from the selected principal component sample set again until the target travel time range is obtained.

[0068] Step S23 specifically involves selecting 11 feature parameters—acceleration time, deceleration time, constant speed time, maximum speed, average speed of the travel segment, standard deviation of speed of the travel segment, standard deviation of acceleration of the travel segment, maximum acceleration, average acceleration of the acceleration segment, minimum deceleration, and average deceleration of the deceleration segment—to participate in the principal component operation. Before the principal component operation, each sample is centered according to the following formula:

[0069]

[0070] Where: μ is the statistical mean of the sample set, σ is the statistical standard deviation of the sample set, x is the feature value of the short-range segment sample, z is the feature value of the short-range segment sample after alignment, k represents the sample number, and i represents the feature parameter number;

[0071] Randomly select several samples from the principal component samples as seed points; iteratively calculate the convergence position using the following formula:

[0072]

[0073] Where: S represents the principal component sample dataset, s represents the sample vector in the principal component sample dataset, T represents the seed point dataset, and t represents the seed point coordinate vector. Let m(T) represent the updated seed point position coordinate vector, and m(T) represent the set of all updated seed point position coordinate vectors. The kernel function K is defined as: The weighting coefficient w is defined as:

[0074] The distance norm of the seed point coordinates for the two iterations is calculated using the following formula:

[0075]

[0076] Where: k represents the coordinate dimension, d represents the value under the specific coordinate dimension, old and new represent the previous iteration and the current iteration, respectively, and N represents the total number of dimensions;

[0077] The loop exits when the distance norm is less than the specified deviation, outputting the latest seed point convergence coordinates; otherwise, the seed point dataset is updated (m(T)→T), and the loop returns to the previous iteration to calculate the convergence position and iterate again. The seed point convergence coordinates are then converted into the desired feature parameter values ​​as class features. The conversion formula is as follows:

[0078]

[0079] Where e1, e2, ..., e p Let c represent the coefficient vector of each principal component sample, where μ and σ represent the statistical mean and standard deviation of the principal component sample dataset, respectively. 1k c 2k c pk This represents the convergence coordinates of the k-th seed point, where p represents the number of feature dimensions. This represents the expected feature of the k-th seed point.

[0080] Step S24 specifically involves converting the vehicle speed into a driving state, using the following conversion formula:

[0081] st = ceil(v / DS) + 1

[0082] Where v represents vehicle speed, DS represents the vehicle speed range corresponding to a unit state, the operator ceil(·) represents rounding up, and st represents the state value corresponding to the current vehicle speed;

[0083] The improved Kneser-Ney smoothing algorithm is used to estimate the state transition probability matrix. The estimation formula is as follows:

[0084]

[0085] Where i and j represent the current and next driving states, respectively, c(ij) represents the number of samples corresponding to the ordered binary state combination ij between two adjacent time points, the denominator on the right side of the equals sign counts the number of samples of state i at the current time, D<·> represents the discount factor, γ(i) represents the interpolation coefficient related to state i, and p(j) represents the probability of unary state j. This represents the state transition probability of the ordered binary state combination ij estimated by the algorithm;

[0086] The final estimated state transition probability matrix model is as follows:

[0087]

[0088] In this context, the first index represents the current state value, the second index represents the next state value, and k is the number of states.

[0089] The output random operating condition specifically involves generating random operating conditions that meet accuracy requirements using the Markov chain principle, including the following steps:

[0090] S31: The average idling time of the statistical sample is used as the expected idling time. The sum of acceleration time, deceleration time and constant speed time in the expected characteristic value is calculated to obtain the expected non-idling time. The expected idling time and the expected non-idling time are added together to obtain the expected total operating time.

[0091] S32: Randomly generate driving conditions using the state transition probability matrix;

[0092] S33: Calculate the average relative deviation of the characteristic parameters between the driving condition and the desired characteristic. The calculation formula is as follows:

[0093]

[0094] Where V represents the total number of feature parameters, the selected feature parameters include maximum speed, average speed of the driving segment, standard deviation of speed of the driving segment, standard deviation of acceleration of the driving segment, maximum acceleration, average acceleration of the acceleration segment, minimum deceleration, average deceleration of the deceleration segment, acceleration ratio of the driving segment, deceleration ratio of the driving segment, and constant speed ratio of the driving segment. Among them, the last three time ratio parameters are calculated with corresponding acceleration times. Z represents the statistical mean of the k-th feature parameter of the current class. k This represents the expected mean of the k-th feature parameter of the current component sample;

[0095] S34: Calculate the relative percentage deviation between the total driving time and non-idling time and the expected total time and expected non-idling time;

[0096] S35: Output the operating condition when the average relative deviation of the characteristic parameters and the relative deviation of the duration are both less than 5%; otherwise, return to step S32 to iterate again.

[0097] A development system for driving conditions with a large sample size includes a data acquisition unit, a data processing unit, and a data output unit. The data acquisition unit is data-connected to the data processing unit, and the data processing unit is data-connected to the data output unit. The data acquisition unit is used to collect real vehicle driving data, segment it into short-distance segment samples, and statistically analyze its driving characteristics. The data processing unit is used to analyze and process the driving characteristics of the short-distance segment samples. The data output unit is used to output random driving conditions.

[0098] For 1 year of driving data collection from 10 passenger vehicles, the data collection frequency was 1Hz. After data cleaning such as outlier removal and smoothing filtering, a time-vehicle speed dataset with more than 9.23 million records was obtained. This time-vehicle speed dataset was used as the original real vehicle driving data input to construct typical driving conditions.

[0099] Based on the definition of short trip, the original data was segmented into short trip samples, and short trip samples with actual travel time of less than 20 seconds were removed, resulting in a total of 69,370 short trip samples.

[0100] The characteristics of each short-stroke sample were statistically analyzed, including: idling time, acceleration time, deceleration time, constant speed time, maximum speed, average speed of the driving segment, standard deviation of speed of the driving segment, standard deviation of acceleration of the driving segment, maximum acceleration, average acceleration of the acceleration segment, minimum deceleration, average deceleration of the deceleration segment, acceleration ratio of the driving segment, deceleration ratio of the driving segment, and constant speed ratio of the driving segment. The statistical results of all short-stroke samples were summarized into a short-stroke sample feature matrix, and some results are shown in Table 1.

[0101]

[0102] Table 1

[0103] Principal component analysis was performed using three feature parameters: acceleration time, deceleration time, and constant speed time. The contribution rates of the three principal components were 89.81%, 9.50%, and 0.69%, respectively. The cumulative contribution rate of the first two principal components exceeded 98%, which is sufficient to cover most of the information in the sample space. The principal component sample set was formed by selecting the first two principal components, as shown in Table 2 below:

[0104]

[0105] Table 2

[0106] This embodiment examines short-distance trips with medium to long durations, so the travel time range was selected as [300, 310], which is greater than or equal to 300 seconds and less than or equal to 310 seconds. A total of 166 samples were obtained.

[0107] Randomly select 50 samples, such as Figure 2 As shown, the distribution range and relative density of the randomly selected samples are quite consistent with the spatial distribution of the complete sample.

[0108] The drawn gamma diagram is as follows Figure 3 As shown in the figure, the random scatter distribution is quite close to the diagonal with a slope of 1, which can be considered to follow a normal distribution.

[0109] because Figure 2 The sample distribution shown is relatively uniform, and Figure 3 The scatter distribution shown approximates a normal distribution, and the selected driving time range can be considered reasonable and can be used as a sub-category for working condition construction.

[0110] Short-distance samples with travel times ranging from 300 to 310 seconds were selected. The short-distance sample data is shown in Table 3 below:

[0111]

[0112] Table 3

[0113] Perform principal component analysis on these samples again. Note that the selected feature parameters are acceleration time, deceleration time, constant speed time, maximum speed, average speed of the travel segment, standard deviation of speed of the travel segment, standard deviation of acceleration of the travel segment, maximum acceleration, average acceleration of the acceleration segment, minimum deceleration, and average deceleration of the deceleration segment, totaling 11 feature parameters. Furthermore, before performing the principal component analysis, first calculate the mean and standard deviation of each component, and then calculate the centering formula according to step S23 to center the sample feature values. The final principal component sample dataset is shown in Table 4 below:

[0114]

[0115] Table 4

[0116] From the principal component samples shown in Table 4, approximately 10% of the samples were randomly selected as seed points. The distribution of these seed points is as follows: Figure 4 As shown.

[0117] The convergence threshold of the distance norm was set to 0.001. The mean shift algorithm was used to iterate over the coordinates of these seed points in the principal component sample distribution space shown in Table 4 until the convergence condition was met. The results are as follows: Figure 5As shown in the figure, the final convergence positions of all seed points coincide, indicating that the sample set has only one cluster center, which also verifies the rationality of subdividing the categories by travel time.

[0118] At this point, the cluster center coordinates are defined by the principal components. Using the principal component coefficient matrix, the statistical mean and statistical standard deviation of each feature parameter, the cluster center coordinates are transformed according to the formula in step S23. The results are shown in Table 5 below:

[0119]

[0120] Table 5

[0121] Calculate the state transition probability matrix according to step S24. Select a vehicle speed range of 10 km / h corresponding to each unit state. The final state transition probability matrix is ​​as follows. Figure 6 As shown. It is worth noting that there are still small probability values ​​at positions far from the diagonal, while the probability value calculated using the maximum likelihood estimation method is 0. This shows that the improved Kneser-Ney smoothing algorithm can take into account small probability times and has more perfect randomness.

[0122] The average idling time of the short-stroke sample was 36 seconds. Based on the acceleration time, deceleration time, and constant speed time, the acceleration ratio, deceleration ratio, and constant speed ratio of the driving segment were calculated, and the driving time was calculated to be 304 seconds. Therefore, the expected total driving time is 340 seconds.

[0123] Following step S3, operating conditions are randomly generated using the state transition probability matrix, and the operating condition curve is shown below. Figure 7 As shown in Table 6 below:

[0124]

[0125] Table 6

[0126] The average relative deviation between the constructed operating condition and the expected characteristic value is 3.85%, the deviation of the non-idle time is 1.64%, and the total operating time is 340 seconds, which meets the accuracy requirement of less than 5%. The constructed operating condition can be used as a typical operating condition for the current sub-category.

[0127] The foregoing is a detailed description of the present invention with reference to the accompanying drawings and preferred embodiments, but the present invention is not limited thereto. Based on the essence of the present invention, those skilled in the art can make various equivalent modifications or substitutions to the embodiments of the present invention, and all such modifications or substitutions should be within the protection scope of the present invention. The protection scope of the present invention should be determined by the scope of the claims.

Claims

1. A large sample-oriented driving mode development method, characterized by, Includes the following steps: S1: Collect real vehicle driving data, segment it into short-stroke segment samples, and statistically analyze its driving characteristics; S2: Analyze and process the driving characteristics of short-distance segment samples; S3: Output random operating conditions; Step S2, specifically the analysis and processing of the driving characteristics of the short-distance segment samples, includes the following steps: S21: Select three characteristic parameters—acceleration time, deceleration time, and constant speed time—for principal component analysis; S22: Determine a suitable range of travel times so that the corresponding principal component sample set conforms to a normal distribution; S23: Filter short-trip segment samples according to the determined travel time range, perform principal component analysis again to obtain principal component sample dataset, and use the mean shift algorithm to cluster the principal component samples to obtain cluster center points and convert the coordinates of the cluster center points into expected class feature parameter values; S24: Divide the states according to driving speed, count the number of samples for different state transitions, and use the improved Kneser-Ney algorithm to estimate the state transition probability matrix.

2. The driving cycle development method according to claim 1, wherein The short-stroke segment sample consists of an idle speed segment followed by a driving segment with a final speed of 0, including the following four states: Acceleration state: A driving state with a non-zero speed and an acceleration greater than or equal to 0.15 m / s2; Deceleration state: A driving state with a non-zero speed and an acceleration less than or equal to -0.15 m / s²; Uniform speed state: A driving state with a non-zero velocity and acceleration within the range of [-0.15, 0.15] m / s²; Idle state: The driving state in which the engine is running continuously but the acceleration and vehicle speed are both 0.

3. The development method for large-sample driving conditions according to claim 1, characterized in that, The specific steps for statistically collecting the driving characteristics of short-distance travel segment samples are as follows: statistically analyzing the driving characteristics of each short-distance travel segment sample and constructing a short-distance travel segment sample feature matrix include the following steps: S11: Statistically analyze the characteristics of each short-stroke sample, including: idling time, driving time, acceleration time, deceleration time, constant speed time, maximum speed, average speed of driving segment, standard deviation of speed of driving segment, standard deviation of acceleration of driving segment, maximum acceleration, average acceleration of acceleration segment, minimum deceleration, average deceleration of deceleration segment, acceleration ratio of driving segment, deceleration ratio of driving segment, and constant speed ratio of driving segment. S12: Select short-trip samples, and the actual driving time of each sample after removing the idling time is greater than or equal to 20 seconds; S13: Construct the feature matrix of short-run samples, i.e., the short-run sample dataset, specifically: each row represents a sample, and each column represents a feature parameter.

4. The development method for large-sample driving conditions according to claim 1, characterized in that, Step S21 specifically involves: selecting sample data with three feature parameters (acceleration time, deceleration time, and constant speed time) from the short-stroke segment sample feature matrix for principal component analysis, and then filtering the number of principal components. The filtering rule is that the cumulative contribution rate of the principal components is greater than or equal to 70%, and finally obtaining the filtered principal component sample set.

5. The development method for large-sample driving conditions according to claim 1, characterized in that, Step S22 specifically involves: selecting a travel time range, with the duration increasing from short to long, and the initial span being greater than or equal to 10 seconds; selecting sample data from the selected principal component sample set based on the time range; randomly selecting an appropriate number of samples from the selected samples, with the sample number accounting for 30% of the total, and plotting a gamma diagram accordingly. When the scatter distribution of the sample points is close to a straight line with a slope of 1, the selected time range is accepted; otherwise, the time range span is increased by 5 seconds, and sample data is selected from the selected principal component sample set again until the target travel time range is obtained.

6. The development method for large-sample driving conditions according to claim 1, characterized in that, Step S23 specifically involves selecting 11 feature parameters—acceleration time, deceleration time, constant speed time, maximum speed, average speed of the travel segment, standard deviation of speed of the travel segment, standard deviation of acceleration of the travel segment, maximum acceleration, average acceleration of the acceleration segment, minimum deceleration, and average deceleration of the deceleration segment—to participate in the principal component operation. Before the principal component operation, each sample is centered according to the following formula: in: It is the statistical mean of the sample set. It is the statistical standard deviation of the sample set. These are the feature values ​​of short-stroke segment samples. It refers to the feature values ​​of short-range segment samples after the middle of the journey. Indicates the sample number. Indicates the feature parameter index; Randomly select several samples from the principal component samples as seed points; iteratively calculate the convergence position using the following formula: in: This represents the principal component sample dataset. This represents the sample vector in the principal component sample dataset. Represents the seed point dataset. Represents the seed point coordinate vector. This represents the updated seed point position coordinate vector. The kernel function represents the set of all updated seed point coordinate vectors. Defined as: Weighting coefficients Defined as: ; The distance norm of the seed point coordinates for the two iterations is calculated using the following formula: in: Indicates the coordinate dimension. This represents the value in a specific coordinate dimension. and These represent the previous iteration and the current iteration, respectively. Indicates the total number of dimensions; The loop exits when the distance norm is less than the specified deviation, and the latest seed point convergence position coordinates are output; otherwise, the seed point dataset is updated. Then, return to the convergence position calculated in the above iteration and iterate again, converting the convergence position coordinates of the seed point into the expected feature parameter values ​​as class features. The conversion formula is as follows: in, This represents the sample coefficient vector of each principal component. and Let these represent the statistical mean and standard deviation of the principal component sample dataset, respectively. Indicates the obtained first The convergence coordinates of the seed points Indicates the number of feature dimensions. Indicates the first The expected features of each seed point.

7. The development method for large-sample driving conditions according to claim 1, characterized in that, Step S24 specifically involves converting the vehicle speed into a driving state, using the following conversion formula: in, Indicates vehicle speed. Operator representing the speed range corresponding to a unit state Indicates rounding up. This represents the state value corresponding to the current vehicle speed; The improved Kneser-Ney smoothing algorithm is used to estimate the state transition probability matrix. The estimation formula is as follows: in, and These represent the current and next driving status, respectively. Represents an ordered binary state combination between two adjacent time points. The corresponding number of samples, the denominator on the right side of the equals sign represents the state at the current moment. The number of samples, Indicates the discount factor. Representation and State The relevant interpolation coefficients, Represents a univariate state The probability, Represents the ordered binary state combination estimated by the algorithm. The state transition probability; The final estimated state transition probability matrix model is as follows: In this system, the first index represents the current state value, and the second index represents the next state value. It represents the number of states.

8. The development method for large-sample driving conditions according to claim 1, characterized in that, The output random operating conditions specifically involve generating random operating conditions that meet accuracy requirements using the Markov chain principle, including the following steps: S31: The average idling time of the statistical sample is used as the expected idling time. The sum of acceleration time, deceleration time and constant speed time in the expected characteristic value is calculated to obtain the expected non-idling time. The expected idling time and the expected non-idling time are added together to obtain the expected total operating time. S32: Randomly generate driving conditions using the state transition probability matrix; S33: Calculate the average relative deviation of the characteristic parameters between the driving condition and the desired characteristic. The calculation formula is as follows: in, This represents the total number of feature parameters. The selected feature parameters include maximum speed, average speed over a travel segment, standard deviation of speed over a travel segment, standard deviation of acceleration over a travel segment, maximum acceleration, average acceleration over an acceleration segment, minimum deceleration, average deceleration over a deceleration segment, acceleration ratio over a travel segment, deceleration ratio over a travel segment, and constant speed ratio over a travel segment. Among these, the last three time ratio parameters are calculated from the corresponding acceleration times. Indicates the first of the current class The statistical mean of each feature parameter Indicates the first component sample of the current component. The expected mean of each feature parameter; S34: Calculate the relative percentage deviation between the total driving time and non-idling time and the expected total time and expected non-idling time; S35: Output the operating condition when the average relative deviation of the characteristic parameters and the relative deviation of the duration are both less than 5%; otherwise, return to step S32 to iterate again.

9. A development system for large-sample driving conditions, characterized in that, The system includes a data acquisition unit, a data processing unit, and a data output unit. The data acquisition unit is data-connected to the data processing unit, and the data processing unit is data-connected to the data output unit. The data acquisition unit is used to collect real-vehicle driving data, segment it into short-stroke segment samples, and statistically analyze its driving characteristics. The data processing unit is used to analyze and process the driving characteristics of short-distance segment samples, and the data output unit is used to output random operating conditions.

Images