Vehicle driving condition construction method based on guided Markov evolutionary algorithm

Abstract

The invention discloses an automobile driving condition construction method based on a guided Markov evolution algorithm. According to the method, vehicle driving data are collected and preprocessed through an Internet of Vehicles platform, feature parameters of short-stroke fragments are extracted, dimension reduction processing is performed by adopting principal component analysis, and driving condition categories are divided based on K-means clustering. And further constructing a state transition probability matrix of two parameters of speed and acceleration, generating an initial working condition population, and performing iterative optimization through a genetic algorithm. In the optimization process, a guided updating algorithm is introduced, and two accepting operators are set to update and judge the optimal and worst working conditions respectively, so that the convergence direction of the algorithm is effectively guided, and the precision, efficiency and stability of working condition construction are improved. The method can also be applied to a vehicle durability test system, an emission test system, an energy consumption evaluation system and the like, and is used for simulating a real user driving behavior and generating a representative driving condition sequence.

Description

Technical Field

[0001] This invention relates to the fields of intelligent transportation systems and vehicle engineering, and in particular to a method for constructing vehicle driving conditions based on a guided Markov evolutionary algorithm. Background Technology

[0002] Vehicle driving conditions, as a core technology in vehicle engineering and intelligent transportation systems, are widely used in vehicle emission testing, energy consumption assessment, and structural durability analysis. With the development of vehicle-to-everything (V2X) technology, driving condition construction methods based on big data have gradually become a research hotspot. Among related technologies, a representative driving condition sequence has been constructed through the collaborative operation of the short-pass method, Markov chain modeling, and genetic algorithms. Specifically, this technical system covers the entire process from data acquisition, feature extraction, cluster analysis to driving condition generation and optimization, including key steps such as speed-time curve modeling, state space partitioning, population initialization, selection, crossover, and mutation operations. The short-pass method achieves local feature modeling of the driving condition by dividing it into acceleration, deceleration, constant speed, and idling segments, while Markov chains are used to characterize the temporal transition characteristics of vehicle states, and genetic algorithms improve the global fitting ability of the driving condition through iterative optimization. While existing methods based on statistics and stochastic simulation have improved the accuracy of driving condition construction to some extent, they still suffer from problems such as combinatorial explosion, slow convergence speed, and difficulty in breaking through local optima when processing large-scale data.

[0003] However, existing driving condition construction methods directly employ traditional genetic algorithms for optimization without introducing effective convergence guidance mechanisms. This can lead to unstable convergence due to excessive randomness during iterations, or the optimal solution being replaced in later iterations, thus affecting the accuracy and efficiency of driving condition construction. Furthermore, existing methods often use single kinematic features in feature parameter extraction, failing to fully consider the impact of velocity range distribution on driving condition representativeness, resulting in insufficient fitting ability of the generated driving conditions to the joint velocity-acceleration distribution. In terms of algorithm design, traditional methods typically rely solely on selection, crossover, and mutation operators, lacking dynamic adjustment of the population update strategy, making it difficult to balance global search and local optimization capabilities at different iteration stages. Due to these technical deficiencies, existing driving condition construction methods struggle to generate high-precision, high-stability driving conditions within a limited time when dealing with complex, high-dimensional vehicle network data, limiting their application effectiveness in practical testing and simulation. Summary of the Invention

[0004] The present invention aims to at least partially solve one of the technical problems in the related art.

[0005] With the continuous growth of vehicle operation data, traditional driving condition construction methods have significant shortcomings in terms of accuracy, efficiency, and stability, making it difficult to meet the needs of complex driving condition modeling. Therefore, this invention aims to provide a vehicle driving condition construction method based on a guided Markov evolutionary algorithm to improve the accuracy, convergence speed, and stability of driving condition construction, thereby more realistically reflecting the actual operating characteristics of vehicles in specific regions or user groups.

[0006] Another objective of this invention is to propose a vehicle driving condition construction device based on a guided Markov evolution algorithm.

[0007] To achieve the above objectives, this invention proposes a method for constructing vehicle driving conditions based on a guided Markov evolutionary algorithm, comprising:

[0008] Vehicle driving data is collected through a vehicle-to-everything (V2X) platform, and the driving data is preprocessed to generate a standardized speed-time series.

[0009] The standardized velocity-time series is divided into multiple short-stroke segments, and feature parameters of each short-stroke segment are extracted. The feature parameters include velocity, acceleration and their interval proportion.

[0010] Principal component analysis is performed on the feature parameters to reduce their dimensionality, thereby obtaining the reduced feature parameter matrix.

[0011] Based on the reduced feature parameter matrix, a clustering algorithm is used to classify the short-distance segments to obtain multiple driving condition categories;

[0012] Construct a state transition probability matrix with two parameters: velocity and acceleration. The state transition probability matrix is ​​used to describe the transition relationship between each state.

[0013] An initial working condition population is generated based on the state transition probability matrix, and optimized iteratively using a genetic algorithm to find a better solution based on the initial working condition population. The genetic algorithm includes selection, crossover, and mutation operations.

[0014] In the optimization iteration process, an update algorithm is introduced, which includes two accepting operators, which are used to update the best and worst working conditions in the current population and the previous generation population, respectively.

[0015] In response to the fact that the objective function value of the newly generated operating condition is better than the objective function value of the old operating condition, the new operating condition replaces the old operating condition;

[0016] In response to the fact that the objective function value of the newly generated operating condition is worse than the objective function value of the old operating condition, a decision is made on whether to retain the new operating condition based on the update probability and the discrimination factor.

[0017] The guided update algorithm guides the convergence direction of the optimization process to generate a driving condition sequence that meets the target error requirements.

[0018] Multiple driving condition categories are combined to generate comprehensive driving conditions tailored to different user groups.

[0019] The vehicle driving condition construction method based on guided Markov evolution algorithm in this invention embodiment may also have the following additional technical features:

[0020] In one embodiment of the present invention, the division of the short-stroke segment includes:

[0021] The segment in the speed-time series from one idle moment to the next idle moment is defined as a short-stroke segment;

[0022] The short-stroke segment includes an acceleration segment, a deceleration segment, a constant speed segment, and an idle speed segment;

[0023] The acceleration segment is the driving range in which the acceleration is greater than or equal to a set threshold.

[0024] The deceleration section is a driving range where the acceleration is less than or equal to a set threshold.

[0025] The idle speed range is the driving range in which the vehicle speed is less than or equal to a set threshold.

[0026] The constant speed range is the driving interval where the acceleration is between a set threshold.

[0027] In one embodiment of the present invention, the feature parameters include:

[0028] Average speed, average driving speed, average acceleration during acceleration, average deceleration during deceleration, percentage of acceleration interval, percentage of deceleration interval, percentage of idle speed interval, standard deviation of acceleration, and standard deviation of speed.

[0029] It also includes the percentage of driving time in multiple speed ranges, used to characterize the driving characteristics of different speed ranges.

[0030] In one embodiment of the present invention, the principal component analysis dimensionality reduction method includes:

[0031] The feature parameter matrix is ​​standardized to eliminate dimensional differences;

[0032] Principal components are extracted based on the standardized feature parameter matrix, and the cumulative contribution rate of the principal components exceeds a set threshold.

[0033] The principal components are used to replace the original feature parameters to improve clustering efficiency and working condition construction accuracy.

[0034] In one embodiment of the present invention, the clustering analysis method includes:

[0035] The K-means algorithm is used to cluster the dimensionality-reduced feature parameter matrix;

[0036] The number of clusters K is determined based on the changing trend of the sum of squared errors within clusters (SSE).

[0037] The clustering results are used to classify driving conditions into three categories: low speed, medium speed, and high speed.

[0038] In one embodiment of the present invention, the construction of the state transition probability matrix includes:

[0039] Velocity and acceleration are discretized into multiple states, and state encoding is performed.

[0040] Based on the state code, the number of transitions between each state is counted to generate a state transition count matrix;

[0041] The state transition probability is calculated based on the state transition number matrix, and a state transition probability matrix is ​​generated.

[0042] In one embodiment of the present invention, the genetic algorithm optimization method includes:

[0043] An initial working condition population is randomly generated based on the state transition probability matrix;

[0044] The parent load case is selected based on the fitness value of the load case by selecting the operator;

[0045] State swapping can be performed using the crossover operator without violating the properties of Markov chains;

[0046] Local state mutations can be performed using mutation operators without violating the properties of Markov chains.

[0047] In one embodiment of the present invention, the guided update algorithm includes:

[0048] In each iteration, the best and worst working conditions are extracted from the current population and the previous generation population, respectively.

[0049] The optimal working condition is updated and judged. If the objective function value of the new working condition is better than that of the old working condition, it is directly replaced.

[0050] If the objective function value of the new operating condition is worse than that of the old operating condition, then the decision on whether to retain the new operating condition is based on the update probability and the discrimination factor.

[0051] The worst-case scenario is updated and judged. If the objective function value of the new scenario is better than that of the old scenario, the old scenario is replaced.

[0052] If the objective function value of the new operating condition is worse than that of the old operating condition, then the decision on whether to retain the new operating condition is based on the update probability and the discrimination factor.

[0053] The update probability decreases with the number of iterations to enhance the convergence stability of the algorithm in later stages.

[0054] To achieve the above objectives, another aspect of the present invention proposes a vehicle driving condition construction device based on a guided Markov evolution algorithm, comprising:

[0055] The data acquisition module is used to collect vehicle driving data and generate standardized speed-time series.

[0056] The feature extraction module is used to extract multiple feature parameters from the velocity-time series;

[0057] The feature dimensionality reduction module is used to perform principal component analysis dimensionality reduction processing on the feature parameters;

[0058] The clustering analysis module is used to classify short-stroke segments based on the dimensionality-reduced feature parameters;

[0059] The state modeling module is used to construct the state transition probability matrix with two parameters: velocity and acceleration.

[0060] The genetic algorithm module is used to generate an initial working condition population based on the state transition probability matrix, and to find a better solution based on the initial population;

[0061] The update algorithm module is used to introduce an update algorithm during the optimization iteration process. It includes two accepting operators, which are used to update the optimal and worst working conditions respectively, and guide the convergence direction of the iterative optimization process.

[0062] The target control module is used to set the target accuracy of driving conditions and limit the number of iterations in the optimization process, so as to generate a driving condition sequence that meets the target error requirements within the specified number of iterations.

[0063] The driving condition combination module is used to combine multiple driving condition categories to generate comprehensive driving conditions for the user group.

[0064] The vehicle driving condition construction method and apparatus based on guided Markov evolution algorithm of the present invention effectively solves the shortcomings of traditional methods such as combinatorial explosion, slow convergence speed and excessive randomness by integrating short-run method and Markov chain method, and combining genetic algorithm and guided update algorithm.

[0065] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0066] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0067] Figure 1 This is a flowchart of a method for constructing vehicle driving conditions based on a guided Markov evolution algorithm according to an embodiment of the present invention;

[0068] Figure 2 This is a comparison chart of data before and after preprocessing according to an embodiment of the present invention;

[0069] Figure 3 This is a schematic diagram of a short-stroke segment according to an embodiment of the present invention;

[0070] Figure 4 This is a graph showing the trend of SSE as a function of the number of clusters K according to an embodiment of the present invention;

[0071] Figure 5 This is a graph showing the correlation between vehicle speed and other parameters at different time intervals according to an embodiment of the present invention.

[0072] Figure 6 This is a flowchart of GMEA working condition construction and optimization according to an embodiment of the present invention;

[0073] Figure 7 This is a diagram illustrating the driving conditions according to an embodiment of the present invention;

[0074] Figure 8 This is a comparison diagram of the joint probability density distribution of velocity and acceleration according to an embodiment of the present invention;

[0075] Figure 9 This is a diagram showing the variation of the minimum error in the working condition construction of GMEA and MCE according to an embodiment of the present invention;

[0076] Figure 10 This is a structural diagram of a vehicle driving condition construction device based on a guided Markov evolution algorithm according to an embodiment of the present invention. Detailed Implementation

[0077] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0078] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0079] The following describes, with reference to the accompanying drawings, a method and apparatus for constructing vehicle driving conditions based on a guided Markov evolutionary algorithm, according to an embodiment of the present invention.

[0080] Figure 1 This is a flowchart of a method for constructing vehicle driving conditions based on a guided Markov evolutionary algorithm according to an embodiment of the present invention, as shown below. Figure 1 As shown, it includes:

[0081] S1, collect vehicle driving data through the vehicle network platform, and preprocess the driving data to generate a standardized speed-time series;

[0082] S2, the standardized velocity-time series is divided into multiple short-stroke segments, and the feature parameters of each short-stroke segment are extracted. The feature parameters include velocity, acceleration and their interval proportion.

[0083] S3, perform principal component analysis to reduce the dimensionality of the feature parameters to obtain the dimensionality-reduced feature parameter matrix;

[0084] S4. Based on the reduced feature parameter matrix, a clustering algorithm is used to classify the short-distance segments to obtain multiple driving condition categories;

[0085] S5, construct a state transition probability matrix with two parameters: velocity and acceleration. The state transition probability matrix is ​​used to describe the transition relationship between each state.

[0086] S6. An initial working condition population is generated based on the state transition probability matrix, and optimized iteratively using a genetic algorithm to find a better solution based on the initial population. The genetic algorithm includes selection, crossover, and mutation operations.

[0087] S7. In the optimization iteration process, an update algorithm is introduced. The update algorithm includes two accepting operators, which are used to update the best and worst working conditions in the current population and the previous generation population, respectively.

[0088] S8, in response to the objective function value of the newly generated working condition being better than the objective function value of the old working condition, the new working condition replaces the old working condition;

[0089] S9, in response to the objective function value of the newly generated working condition being worse than the objective function value of the old working condition, determine whether to retain the new working condition based on the update probability and the discrimination factor;

[0090] S10, the update algorithm guides the convergence direction of the optimization process to generate a driving condition sequence that meets the target error requirements;

[0091] S11 combines multiple driving condition categories to generate comprehensive driving conditions tailored to the user group.

[0092] Specifically, the method for constructing vehicle driving conditions based on the guided Markov evolution algorithm according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings:

[0093] In one embodiment of the present invention, in order to obtain actual driving data of the vehicle, an on-board terminal and a vehicle networking platform are used to collect actual driving data. In order to truly reflect the actual road driving characteristics, a refresh rate of 1Hz is often used to describe the transient characteristics of the vehicle operation when designing the working conditions. The sampling frequency of the device is 0.5Hz, so the data needs to be resampled. Since the driving conditions are mainly composed of speed-time curves, the core parameters mainly include speed and time parameters. By collecting driving data of 18 passenger cars for one month, a total of about 1.57 million sets of driving data were collected. Due to the influence of factors such as communication network quality and acquisition hardware conditions, there may be data anomalies caused by signal discontinuity, long-term idling and violent acceleration and deceleration during the data acquisition process. Therefore, it is necessary to preprocess the speed time series extracted from the data, remove abnormal or invalid data and generate new speed time series. The main components include: (1) treating congested road sections as idling; (2) treating acceleration greater than 3m·s -2 or less than -3 m·s -2 (3) Smoothing is performed on periods of abnormal acceleration; (4) Idle time is not greater than 180s, etc. The processed data is used as the overall data, and its velocity and acceleration distribution is as follows: Figure 2 As shown. Figure 2 In the diagram, (a) represents the velocity distribution. Figure 2 (b) Acceleration distribution in the figure.

[0094] A short-stroke segment is a speed-time segment that begins at one idle moment and ends at the next idle moment. It mainly includes four types of driving segments: acceleration, deceleration, constant speed, and idling. Figure 3 As shown, the acceleration segment is defined as an acceleration greater than or equal to 0.15 m / s². -2 The driving range includes a deceleration section where the acceleration is less than or equal to -0.15 m / s². -2 The driving range includes an idle speed range of less than or equal to 0.5 m / s. -1 The driving range is characterized by a constant speed with an acceleration greater than -0.15 m / s². -2And less than 0.15 m·s -2 The driving range was used to extract approximately 18,000 short-trip segments.

[0095] Extracting reasonable characteristic parameters of driving conditions is a prerequisite for designing vehicle driving conditions. This invention extracts nine commonly used characteristic parameters to characterize each kinematic segment, and the definitions of each parameter refer to GB / T 38146.1-2019. These mainly include: average speed, average driving speed, average acceleration during acceleration, average deceleration during deceleration, percentage of acceleration interval, percentage of deceleration interval, percentage of idle speed interval, standard deviation of acceleration, and standard deviation of speed. Furthermore, to improve the accuracy of the speed-acceleration probability distribution and clustering effect of the design conditions, this invention also introduces nine characteristic parameters characterizing the percentage of speed intervals. All characteristic parameters are shown in Table 1, and the extracted characteristic parameter matrix A... m×n As shown in equation (1):

[0096] Table 1 Feature Parameter Extraction Set

[0097] Parameter name symbol unit Parameter name symbol unit average speed <![CDATA[v p ]]> <![CDATA[m·s -1 ]]> Percentage of speeds in the 0-10km / h range <![CDATA[P 0-10 ]]> ％ Average driving speed <![CDATA[v sp ]]> <![CDATA[m·s -1 ]]> Percentage of speeds in the 10-20km / h range <![CDATA[P 10-20 ]]> ％ Average acceleration during acceleration phase <![CDATA[a ac ]]> <![CDATA[m·s -2 ]]> Percentage of speeds in the 20-30km / h range <![CDATA[P 20-30 ]]> ％ Average deceleration during deceleration phase <![CDATA[a de ]]> <![CDATA[m·s -2 ]]> Percentage of speeds in the 30-40km / h range <![CDATA[P 30-40 ]]> ％ Acceleration ratio <![CDATA[P ac ]]> ％ Percentage of speeds in the 40-50km / h range <![CDATA[P 40-50 ]]> ％ Deceleration ratio <![CDATA[P de ]]> ％ Percentage of speeds in the 50-60km / h range <![CDATA[P 50-60 ]]> ％ Idle speed ratio <![CDATA[P di ]]> ％ Percentage of speeds in the 60-70km / h range <![CDATA[P 60-70 ]]> ％ Acceleration standard deviation during acceleration phase <![CDATA[a std ]]> <![CDATA[m·s -2 ]]> Percentage of speeds in the 70-80km / h range <![CDATA[P 70-80 ]]> ％ speed standard deviation <![CDATA[v std ]]> <![CDATA[m·s -1 ]]> Percentage of speeds of 80km / h and above <![CDATA[P 80+ ]]> ％

[0098]

[0099] In the formula, a ij (i = 1, ..., m; j = 1, ..., n) represents the j-th type of feature parameter for the i-th segment. m is the number of short-stroke segments, and n is the type of feature parameter.

[0100] Furthermore, excessively high feature dimensionality can negatively impact clustering efficiency and results. Reducing feature dimensionality can effectively improve clustering accuracy. Principal Component Analysis (PCA) is used to reduce the dimensionality of features, extracting principal components from the data and replacing the original features with these principal components to improve model efficiency. The main steps of dimensionality reduction are as follows:

[0101] By standardizing the data, the influence of different units can be reduced, thereby reducing the error bias of the results. The formula for standardizing feature data is shown in equation (2), and the standardized feature matrix Z m×n As shown in equation (3):

[0102]

[0103] In the formula, z ij (i = 1, ..., m; j = 1, ..., n) are the standardized feature parameters, μ j Let σ be the mean of the j-th feature of all segments. j Let be the standard deviation of the j-th feature of all segments.

[0104] Analysis of the characteristic parameters of each short-stroke segment revealed that the speed of most of these segments was less than 70 km / h. -1 Therefore, P can be ignored. 70-80 and P 80+ The impact of these two features on PCA dimensionality reduction. During dimensionality reduction, only the first 16 feature parameters of the feature matrix were selected. MATLAB's PCA function was used to perform dimensionality reduction on the standardized feature parameter matrix. The contribution rates of each principal component and the cumulative contribution rate are shown in Table 2. When the cumulative contribution rate of the principal components exceeds 85%, the contribution rate of subsequent components decreases rapidly, indicating that the current principal components can reflect most of the features of the data before dimensionality reduction. Therefore, the first 6 principal components were extracted as the feature values ​​of the short-run segment. After principal component analysis feature dimensionality reduction, the 16-dimensional feature parameter matrix was reduced to 6 dimensions. The dimensionality-reduced feature matrix is ​​shown in Table 3.

[0105] Table 2 Explanation of Total Variance

[0106]

[0107]

[0108] Table 3 Feature Matrix After Dimensionality Reduction

[0109]

[0110] K-means clustering is a simple and widely used clustering algorithm, mainly used to divide points in a dataset into K clusters, minimizing the distance between each point and the center of its cluster. By setting a reasonable K value, short-distance segments with high similarity can be clustered together, thus reconstructing the working conditions. In MATLAB, clustering is performed based on the K-means function. The sum of squared errors within clusters (SSE) is one of the important indicators used to evaluate the clustering effect in cluster analysis. It measures the sum of the average distances between data points within each cluster and the cluster center. The calculation formulas are shown in equations (4) and (5). The SSE variation trend of the feature matrix after dimensionality reduction and clustering with different numbers of clusters is shown in equation (5). Figure 4 As shown.

[0111]

[0112] In the formula, K is the total number of clusters, and Y ik (i = 1, ..., m; k = 1, ..., N) is the k-th feature of the i-th data set, C jk It is the k-th feature of the j-th cluster center, and N is the number of features in each group of data.

[0113] As the value of K increases, the SSE gradually decreases because more clusters can better adapt to the data. When K is greater than or equal to 3, the decrease in SSE is significantly reduced. To balance clustering accuracy and computational efficiency, clustering analysis was performed with K=3, resulting in three types of short-distance segments. By concatenating the short distances within each cluster end-to-end, three types of driving data were reconstructed. Based on the average speed difference of the driving data in each cluster after clustering, they were named low-speed, medium-speed, and high-speed driving conditions, respectively. The statistical characteristic parameters of the three driving conditions are shown in Table 4.

[0114] Table 4 Characteristic parameters for three types of driving conditions

[0115] feature Low-speed operating conditions Medium-speed operating conditions High-speed operating conditions feature Low-speed operating conditions Medium-speed operating conditions High-speed operating conditions <![CDATA[v p ]]> 1.510 5.347 12.638 <![CDATA[P 0-10 ]]> 77.051 28.628 13.763 <![CDATA[v sp ]]> 4.759 6.729 14.183 <![CDATA[P 10-20 ]]> 13.749 22.549 5.615 <![CDATA[a ac ]]> 0.538 0.463 0.466 <![CDATA[P 20-30 ]]> 5.201 24.690 8.014 <![CDATA[a de ]]> -0.454 -0.458 -0.495 <![CDATA[P 30-40 ]]> 2.478 17.232 12.269 <![CDATA[P ac ]]> 11.613 27.753 26.813 <![CDATA[P 40-50 ]]> 1.074 5.727 16.060 <![CDATA[P de ]]> 10.419 25.548 24.726 <![CDATA[P 50-60 ]]> 0.361 0.946 15.107 <![CDATA[P di ]]> 68.263 20.545 10.892 <![CDATA[P 60-70 ]]> 0.063 0.167 11.637 <![CDATA[a std ]]> 0.394 0.285 0.299 <![CDATA[P 70-80 ]]> 0.019 0.042 7.311 <![CDATA[v std ]]> 2.779 3.892 7.566 <![CDATA[P 80+ ]]> 0.005 0.020 10.224

[0116] A Markov chain is a stochastic process in a state space that transitions from one state to another. This process requires the "memoryless" property: the probability distribution of the next state is determined solely by the current state, and events preceding it in the time series are irrelevant. Conducting Markov property tests is fundamental to using the Markov chain method for constructing work conditions. Correlation analysis was performed on vehicle speeds in the datasets before and after clustering at different time intervals. The results are as follows... Figure 5 As shown, when the time interval is 1 second, the correlation between vehicle speeds is close to 1. As the time interval gradually increases, the correlation between vehicle speeds decreases rapidly. Therefore, it can be considered that the vehicle speed at the current moment is only related to the vehicle speed in the previous second, proving that the vehicle speeds in the datasets before and after clustering satisfy the Markov property.

[0117] Based on short-trip segment clustering, a Markov chain method is used to generate the optimal driving condition for each class. Finally, the driving conditions of each class are combined to obtain the user's driving conditions. A guided Markov evolutionary algorithm is proposed to generate the optimal driving conditions, integrating genetic algorithms and update algorithms. The process is as follows: Figure 6 As shown, the algorithm comprises three parts: state transition matrix construction, genetic algorithm, and update algorithm. Within this framework, the state transition matrix constructs a population of working conditions that conforms to Markov properties. The genetic algorithm iteratively evolves the population, expanding the search range to gradually approach the target solution. The core of the guided Markov evolutionary algorithm is the update algorithm. Two types of acceptance operators, designed based on the characteristics of the working condition construction, enable the algorithm to accept inferior solutions with a certain probability in the early stages of iteration to avoid getting trapped in local optima. In the later stages of iteration, they ensure that the algorithm retains the optimal solution with a high probability, thus guiding the algorithm to converge and ensuring that the optimization process steadily progresses towards the desired goal. This design not only retains the global search capability of the genetic algorithm but also enhances its local search capability.

[0118] A state transition probability matrix was constructed using two parameters, velocity and acceleration. Shi et al. proved that the two-parameter state transition probability matrix still satisfies the Markov property. Markov chains require discrete states; therefore, state partitioning and encoding of the operating conditions are necessary. The range and step size of velocity and acceleration are determined from the overall data, with the velocity range being 0 to v. max The acceleration range is a min to a max When the step size is too large, the accuracy of the working condition construction is poor; when the step size is too small, the Markov property of the data will be violated and the algorithm's computational efficiency will be low. To balance computational efficiency and accuracy, the speed step size Δv is set to 0.5 m·s. -1 The acceleration step size Δa is equal to 0.5 m·s. -2 Finally, the state is encoded.

[0119] velocity v at time t t Status code m t :

[0120]

[0121] acceleration a at time t t State code n t :

[0122]

[0123] For the state transition probability matrix with two parameters, velocity and acceleration, the joint state code s at time t is... t for:

[0124] s t =m t +(n t -1)×M (8)

[0125]

[0126] In the formula, M is the number of speed-coded states.

[0127] At each step of a Markov chain, the system can change from one state to another or remain in the current state. Both changing and remaining in the state are recorded as a state transition. The matrix composed of all the state transitions is the state transition matrix, as shown in equation (10):

[0128]

[0129] In the formula, S ij (i = 1, ..., q; j = 1, ..., q) represents the number of times a state transitions from the i-th state to the j-th state, and q represents the number of state codes.

[0130] The probability of changing or maintaining a state is called the state transition probability. State transition probability P ij The calculation formula is shown in equation (11):

[0131]

[0132] In the formula, P ij (i = 1, ..., n; j = 1, ..., n) represents the probability of transitioning from the i-th state to the j-th state.

[0133] The matrix composed of all state transition probabilities is the state transition probability matrix, as shown in equation (12):

[0134]

[0135] A certain number of working conditions are randomly generated based on the state transition probability matrix as the initial population. The number of individuals in each working condition population is initially set to 100. The calculation formula for the state code based on the state transition probability matrix is ​​shown in equation (13):

[0136]

[0137] In the formula, P ij s' represents the probability of transitioning from state i to state j. t This represents the sequence number of the next state code, where r is a random number between 0 and 1. (This is achieved through s'...) t The corresponding state code can be obtained, and the speed code can be calculated from the state code. The speed code represents the speed value within a certain range, and the speed code needs to be decoded using equation (14):

[0138] v p =(m' t -1)×Δv+r×Δv (14)

[0139] In the formula, v p The predicted velocity value, m' t This represents the speed state code corresponding to the state code after decoding, where r is a random number between 0 and 1, and Δv is the speed step size.

[0140] Referring to test conditions such as CLTC and WLTC, and considering the integrity of kinematic segments, the total length of the driving condition is set to 1800s, and the set length L of the speed sequence for each condition is... I The length L of the velocity sequence for each type of working condition after clustering J The proportion of the total data velocity sequence length L is determined, as shown in equation (15):

[0141]

[0142] The objective function is used to evaluate the performance of an individual in the algorithm and determines the convergence direction of the algorithm. To solve the multi-objective convergence optimization problem, an objective function F is designed based on the proposed multi-feature unification function for Markov evolution algorithms, as shown in equation (16):

[0143]

[0144] In the formula, d represents the number of evaluation indicators, and X represents the generated predicted operating conditions.

[0145] Among them, H j (X) is the error judgment function, and its calculation formula is shown in equation (17):

[0146]

[0147] In the formula, f i (X) represents the absolute value of the error in calculating the evaluation index. It is a vector of allowable values ​​for the absolute value of the relative error of a single evaluation index, representing the target accuracy for a single type of working condition.

[0148] f i The formula for calculating (X) is shown in equation (18):

[0149]

[0150] In the formula, I ei (X) is the i-th feature parameter used to construct the operating condition sequence, I oi Let be the i-th feature parameter of the overall data.

[0151] The objective function F reflects the magnitude of the error of the constructed operating condition sequence relative to the original data. By setting the error judgment function, it can be ensured that the errors of each feature gradually converge to the allowable value.

[0152] The selection process determines which individual in the population will be used for reproduction and to produce the next generation. Work condition sequences with lower errors are more likely to be selected, but work condition sequences with higher errors still have a chance of being selected, in order to avoid the algorithm getting stuck in local optima during the iteration process. The selection operator based on the roulette wheel selection method is shown in equation (19):

[0153]

[0154] In the formula, Y i Let p(X) be the index of the work condition sequence in the original population, p(X) be the probability of a single work condition sequence being selected, and r be the index of the work condition sequence in the original population. i The number is randomly generated between 0 and 1, and n is the number of work condition sequences in the original population. A new population is formed by finding the corresponding work condition sequence from the original population according to its index for subsequent operations.

[0155] To increase the probability of selecting working conditions with lower errors, thereby further improving the accuracy and efficiency of subsequent evolution, an exponential probability calculation method was designed based on the error characteristics of the working condition sequence. The calculation formula for the selection probability function p(X) is shown in equation (20):

[0156]

[0157] In the formula, F(X) represents the fitness value corresponding to the Xth working condition, and F(Y) represents the fitness value corresponding to the Yth working condition.

[0158] Crossover involves exchanging chromosomes between two selected parent sequences, and the resulting work case sequence should strictly follow the rules of a Markov chain. If chromosome crossover is performed directly based on the corresponding positions of the work cases, the randomness of chromosome selection can easily disrupt the Markov property of the crossover work cases. Therefore, a crossover operator that satisfies the Markov property should be designed.

[0159] First, two working condition sequences are randomly selected from the chosen population, as shown in equations (21) and (22):

[0160]

[0161] In the formula, This is the state code corresponding to the velocity and acceleration parameters per second in the operating condition sequence, where n is the length of a single operating condition sequence.

[0162] Then, chromosomes of equal length are extracted from the two sequences, as shown in equations (23) and (24):

[0163]

[0164]

[0165] In the formula, i≥1,j≤n-1.

[0166]

[0167] If two chromosomes satisfy the requirements of equation (25), then an exchange can occur between the two chromosomes, transferring the first sequence to the second chromosome. With the second sequence The crossover operation is completed by exchanging data. During the crossover process, the population is divided into two equal groups. Each time, one work case is selected from each group for crossover. The selected work case sequence is not repeated during the crossover operation.

[0168] Mutation allows algorithms to explore a wider solution space, helping to discover more potential solutions and increasing the probability of finding the global optimum. Mutation is performed on the population without violating the Markov property of the operating conditions. The mutated individuals are randomly selected proportionally from the population, with a mutation probability of 10%. The mutation algorithm randomly selects a point in the operating condition sequence for mutation. Starting from the state code corresponding to that point, it regenerates the operating condition with the terminating state of idling using the state transition probability matrix, and the length of the operating condition is the same as the length of the operating condition before mutation.

[0169] Traditional genetic algorithms possess strong global search capabilities, guiding the population towards a better outcome through continuous evolution. However, in the later stages of iteration, the differences in error between different task sequences within the population decrease. Due to the randomness of selection, crossover, and mutation operators, the probability of the best individual being eliminated gradually increases, making it difficult for the algorithm to converge to the target accuracy in the later stages of evolution. Furthermore, due to the specific nature of task construction, the similarity of task sequences within the population gradually increases in the later stages of iteration, making it difficult to provide new evolutionary directions during crossover and mutation, severely impacting the algorithm's efficiency and accuracy.

[0170] Therefore, this invention designs an update algorithm based on the characteristic that the construction of driving conditions must obey the Markov property. The update algorithm extracts two pairs of solutions, the best and the worst, from the current population and the previous generation, respectively. Then, it performs update judgments on the best and worst solutions. If the current population's solution is better than the previous generation's solution, the corresponding solution is directly updated to the new solution; otherwise, the update probability is used to determine whether to update. By introducing the update algorithm, not only is the global search capability of the genetic algorithm retained, but the local search capability of the algorithm is also enhanced through two judgment processes, effectively guiding the convergence direction of the algorithm. The update judgment for the best driving condition ensures that the algorithm accepts the worst solution with a certain probability in the early stage to avoid getting trapped in local optima, and ensures that the algorithm retains the best solution with a high probability in the later stage, thus achieving stable convergence. The update judgment for the worst driving condition ensures that the population gradually moves towards a better solution while the best solution remains unchanged, avoiding getting trapped in local optima. The update algorithm updates only the best and worst conditions in the original population instead of replacing the entire population. This reduces the update frequency of individuals in the population, thereby increasing the diversity among individuals in the later stages of iteration and providing more evolutionary directions in subsequent evolutionary processes. The main pseudocode of the update algorithm is shown in Table 5:

[0171] Table 5. Update Algorithm Pseudocode

[0172]

[0173] The discriminant factor R is a random number between 0 and 1, and the update probability P T The calculation formula is shown in equation (26).

[0174]

[0175] The design of the attenuation number T is shown in equation (27):

[0176] T i+1 =0.8T i (27)

[0177] In the formula, T i This is the decay number corresponding to the current generation, with the initial decay number T1 being 1000.

[0178] The GMEA algorithm was used to construct and optimize driving conditions, with a maximum iteration limit of 500. Due to the varying speed ranges and sequence lengths of the driving conditions, the evaluation metrics for low-speed conditions were set to the first 13 characteristic parameters, with a relative error tolerance of 5%; for medium-speed conditions, the evaluation metrics were set to the first 15 characteristic parameters, with a relative error tolerance of 1%; and for high-speed conditions, the evaluation metrics were set to the first 15 characteristic parameters, with a relative error tolerance of 2%. Using MATLAB's built-in timer, the computation time was 172.8 seconds. A total of 1800 seconds of driving conditions were constructed, including a 360-second low-speed condition, a 509-second medium-speed condition, and a 931-second high-speed condition. The constructed driving conditions are shown below. Figure 7 As shown.

[0179] The characteristic parameters of the construction test data and the overall data are compared in Table 6. The accuracy of the construction test data is evaluated by the mean absolute percentage error (MAPE), and the formula for calculating MAPE is shown in equation (28):

[0180]

[0181] In the formula, δ i t represents the percentage error of the i-th feature parameter. i Let t' be the i-th feature parameter of the overall data. i To construct the i-th characteristic parameter of the working condition, N f The number of features involved in the calculation.

[0182] The rationality of the constructed driving condition was determined by comparing the 18 characteristic parameters described in section 1.2. Table 6 shows that the MAPE of the constructed driving condition is only 1.06%, and the maximum percentage error of a single characteristic parameter is only 2.54%, indicating that the error in the constructed driving condition is small and stable. The results show that the constructed driving condition can truly reflect the actual kinematic driving characteristics of the vehicle. To verify the stability of the method and the repeatability of the results, the driving condition was constructed 10 times, and the results are shown in Table 7. The construction time for each driving condition was controlled within 200 seconds, and the average time for 10 times was 172.8 seconds. The percentage error of each characteristic in the 10 driving condition constructions was less than 5%, and the MAPE of the 10 constructed driving conditions was less than 2%, with the minimum MAPE being only 0.79%, indicating that the method can stably construct high-precision driving conditions.

[0183] Table 6 Comparison of Feature Parameters

[0184]

[0185]

[0186] Table 7. Relative errors and computation times of various characteristic parameters in ten operating scenarios.

[0187]

[0188] The overall data and the velocity-acceleration joint probability density distribution (VA matrix) of the built-in operating conditions are as follows: Figure 8 As shown. Figure 8 In the diagram, (a) represents the joint probability density distribution of the overall data. Figure 8 (b) in the figure represents the joint probability density distribution of the constructed working conditions. The distribution trends of the VA matrices of the constructed working conditions and the overall data are almost identical. The formula for calculating the similarity coefficient is shown in Equation (30), and the formula for calculating the average distribution error is shown in Equation (31). The similarity coefficient of the VA matrices of the two reaches 99.71%, and the distribution errors of the two are mainly located between -1 and 1 m·s. -2 Within the acceleration range, the distribution error of each interval is less than 1.1%, and the average distribution error is only 0.17%, indicating that the constructed working condition can well reproduce the joint probability density distribution of the overall data.

[0189]

[0190] In the formula, Ω is the similarity coefficient, and B pi Let C be the i-th component of the one-dimensional permutation of the original data VA matrix. pi To construct the i-th component of the one-dimensional arrangement of the VA matrix for the operating conditions, z' represents the total number of components in the one-dimensional arrangement of the VA matrix. η is the error, and H... j To construct the proportion of the j-th distribution interval of the operating condition, h jdenoted as , where is the proportion of the j-th distribution interval in the original data, and v is the number of distribution intervals. Table 8 shows the error of each interval in the joint probability density distribution of velocity and acceleration.

[0191] Table 8 Errors in different intervals of the joint probability density distribution of velocity and acceleration.

[0192]

[0193] To verify the impact of the update algorithm on accuracy in GMEA, a comparative test was conducted using the Ordinary Markov Evolutionary Algorithm (MCE) without an update algorithm. MCE only used a genetic algorithm for task construction and optimization. The data preprocessing procedures, extracted feature parameters, and allowable values ​​for evaluation metrics were kept consistent for both algorithms. The error variation of each iteration for both algorithms was as follows: Figure 9 As shown, under the same number of iterations, the errors of the final working cases constructed by GMEA and MCE were 1.03% and 2.21%, respectively. During the iteration process, GMEA could stably converge towards the target error in the later stages of iteration, while MCE had poorer convergence performance, and both showed some degree of convergence failure in the later stages of iteration. The results show that the update algorithm can significantly improve the accuracy of the constructed working cases and effectively control the convergence direction of the algorithm. Figure 9 (a) Low-speed operating condition, Figure 9 (b) Medium-speed operating condition, Figure 9 (c) High-speed operating condition.

[0194] To further verify the efficiency of the update algorithm in the GMEA algorithm, a comparison was made with the MCE algorithm. When the allowable value of the evaluation index is small, the algorithm struggles to reach the allowable value within the limited number of evolutions. To analyze the efficiency changes of the update algorithm, the allowable value of all evaluation indices was changed to 5%. Ten cycles of work case construction tests were repeated using both methods, and the results are shown in Table 9. The GMEA algorithm has a shorter average computation time, only 47.2% of that of the MCE algorithm, and the average error over 10 cycles is reduced by 1.4% compared to the MCE algorithm. These results demonstrate that by introducing the update algorithm, the computational efficiency of the algorithm can be significantly improved while maintaining work case accuracy.

[0195] Table 9 Comparison of GMEA and MCE operating condition constructions

[0196]

[0197] To verify the impact of speed interval parameters on the construction of work cases, ten work case constructions were performed using the GMEA method with 9 feature parameters (excluding speed interval feature parameters) and 18 feature parameters, respectively. The results are shown in Table 10. Here, the 9-parameter error represents the MAPE of the constructed work case with 9 feature parameters, and the 18-parameter error represents the MAPE of the constructed work case with 18 feature parameters. Table 10 shows that when constructing work cases with 18 feature parameters, both the 9-parameter error and the 18-parameter error are lower, and the VA similarity coefficient is higher, with the average VA similarity coefficient increasing by 1.27%. The results indicate that work cases constructed based on 9 feature parameters are difficult to fit the VA distribution of the original data. Introducing an additional 9 speed interval feature parameters can effectively improve the VA distribution of the constructed work cases, increase the VA similarity coefficient of the constructed work cases, reduce the error of the constructed work cases, and optimize the accuracy and stability of the algorithm to a certain extent.

[0198] Table 10 Comparison of Construction Conditions with Different Numbers of Characteristic Parameters

[0199]

[0200] To verify the applicability of the GMEA method, the build cycle 3.1 was compared with the China Passenger Vehicle Test Cycle (CLTC-P), the Worldwide Harmonized Light Vehicles Test Cycle (WLTC), the European Endurance Standard Test Cycle (NEDC), and the FTP75 cycle. The characteristic parameters and errors of each cycle are shown in Table 11, where P... c The scale represents the uniform speed driving condition. The average errors between the constructed driving condition, NEDC, FTP75, WLTC, and CLTC-P driving conditions and the overall data are 1.14%, 19.73%, 24.01%, 28.90%, and 11.06%, respectively. The results show that the characteristics of the regulatory test conditions differ significantly from the actual driving data of vehicles in a specific area, while the constructed driving conditions can accurately reproduce the actual driving characteristics, indicating that this method can be effectively used to construct high-precision driving conditions for users.

[0201] Table 11 Comparison of characteristics of various regulatory conditions and construction conditions with overall data.

[0202]

[0203] (1) By combining the short-stroke method and the Markov chain method to construct driving conditions, the dataset can be classified according to the feature parameters. The driving conditions are constructed based on the two-parameter state transition probability matrix, which can fully reflect the randomness and instantaneous driving characteristics of driving conditions, and provide a basis for analyzing various driving conditions and constructing high-precision driving conditions. A guided Markov evolution algorithm for constructing user driving conditions is proposed. By introducing the design of the update algorithm and using the design of the acceptor operator to guide the algorithm iteration direction, a new idea is provided for controlling the convergence direction of the algorithm.

[0204] (2) Based on actual passenger vehicle driving data, the error of each characteristic parameter of the constructed working condition can be controlled within 5%, MAPE can be controlled within 2%, the error of the speed-acceleration joint probability density distribution is only 0.17%, and the calculation time is controlled within 190 seconds. Compared with the regulatory test conditions in different regions, the error of each characteristic and the average error of the constructed working condition are smaller than those of the regulatory working condition, indicating that the proposed method can construct a working condition that matches the actual driving conditions of users and can be used as one of the methods for high-precision working condition design.

[0205] (3) When setting the objective function of multi-feature integration, the calculation method of equal feature weights was adopted, without considering the weight of each feature in the iteration process, which resulted in a large error in the feature parameters of the constructed working condition and the construction time of the working condition was unstable. The next step will be to explore the weight evaluation and calculation method of each feature parameter and analyze the impact of the weight on the accuracy and efficiency of the method.

[0206] This invention constructs driving conditions by combining the short-stroke method and the Markov chain method. It introduces speed range features to improve clustering and optimize the joint speed-acceleration distribution of the constructed conditions. The selection operator design is optimized, and a Guided Markov Evolutionary Algorithm (GMEA) is proposed for constructing and optimizing driving conditions. GMEA incorporates an update algorithm within the traditional genetic algorithm framework. This update algorithm mainly includes two acceptor operators. By judging the optimal and worst conditions before and after evolution, it guides the convergence direction of the algorithm, effectively avoiding problems such as insufficient accuracy and slow convergence speed caused by excessive randomness in the iteration process. This significantly improves the efficiency and accuracy of driving condition construction. First, real vehicle driving big data is obtained through a vehicle networking platform. Second, the short-stroke clustering method is used to distinguish different driving conditions. Then, driving condition construction and optimization are studied based on the GMEA algorithm. Finally, the results are compared and analyzed with actual driving data and existing driving condition design methods to verify the effectiveness and superiority of the proposed method.

[0207] To achieve the above embodiments, such as Figure 10 As shown, this embodiment also provides a vehicle driving condition construction device 10 based on a guided Markov evolution algorithm, including:

[0208] Data acquisition module 100 is used to collect vehicle driving data and generate standardized speed-time series;

[0209] Feature extraction module 200 is used to extract multiple feature parameters from the velocity-time series;

[0210] The feature dimensionality reduction module 300 is used to perform principal component analysis dimensionality reduction processing on the feature parameters;

[0211] Clustering analysis module 400 is used to classify short-stroke segments based on the dimensionality-reduced feature parameters;

[0212] State modeling module 500 is used to construct a state transition probability matrix with two parameters: velocity and acceleration.

[0213] The genetic algorithm module 600 is used to generate an initial working condition population based on the state transition probability matrix, and to find a better solution based on the initial population.

[0214] The update algorithm module 700 is used to introduce an update algorithm during the optimization iteration process. It includes two accepting operators, which are used to update the optimal working condition and the worst working condition respectively, and guide the convergence direction of the iterative optimization process.

[0215] The target control module 800 is used to set the target accuracy of the driving conditions and limit the number of iterations in the optimization process, so as to generate a driving condition sequence that meets the target error requirements within the specified number of iterations.

[0216] The driving condition combination module 900 is used to combine multiple driving condition categories to generate comprehensive driving conditions for user groups.

[0217] Furthermore, it also includes a velocity-acceleration joint probability density distribution modeling module, used for:

[0218] Construct the joint probability density distribution matrix of velocity and acceleration;

[0219] Calculate the similarity coefficient between the operating condition sequence and the original data in the joint velocity-acceleration distribution;

[0220] The similarity coefficient is used to evaluate the fitting accuracy of the working condition on the joint distribution of velocity and acceleration.

[0221] Furthermore, it also includes an error control module for:

[0222] The matching degree between the working condition sequence and the original data is calculated based on the multi-feature unified error function;

[0223] The error function includes the absolute value of the average relative percentage error of multiple characteristic parameters;

[0224] The error control module is used to determine whether the operating condition meets the target error requirement.

[0225] The device of this invention constructs driving conditions by combining the short-stroke method and the Markov chain method. It introduces speed range features to improve clustering effects and optimize the joint speed-acceleration distribution of the constructed conditions. The design of the selection operator is optimized for constructing and optimizing driving conditions. GMEA incorporates an update algorithm design into the traditional genetic algorithm framework. This update algorithm mainly includes two acceptance operators. By judging the optimal and worst conditions before and after evolution, it guides the convergence direction of the algorithm, effectively avoiding problems such as insufficient accuracy and slow convergence speed caused by excessive randomness in the iteration process, significantly improving the efficiency and accuracy of condition construction. First, real vehicle driving big data is obtained through a vehicle networking platform. Second, the short-stroke clustering method is used to distinguish different driving conditions. Then, the GMEA algorithm is used to construct and optimize driving conditions. Finally, the results are compared and analyzed with actual driving data and existing driving condition design methods to verify the effectiveness and superiority of the proposed method.

[0226] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0227] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A method for constructing vehicle driving conditions based on a guided Markov evolutionary algorithm, characterized in that, Includes the following steps: Vehicle driving data is collected through a vehicle-to-everything (V2X) platform, and the driving data is preprocessed to generate a standardized speed-time series. The standardized velocity-time series is divided into multiple short-stroke segments, and feature parameters of each short-stroke segment are extracted. The feature parameters include velocity, acceleration and their interval proportion. Principal component analysis is performed on the feature parameters to reduce their dimensionality, thereby obtaining the reduced feature parameter matrix. Based on the reduced feature parameter matrix, a clustering algorithm is used to classify the short-distance segments to obtain multiple driving condition categories; Construct a state transition probability matrix with two parameters: velocity and acceleration. The state transition probability matrix is ​​used to describe the transition relationship between each state. An initial working condition population is generated based on the state transition probability matrix, and optimized iteratively using a genetic algorithm to find a better solution based on the initial population. The genetic algorithm includes selection, crossover, and mutation operations. In the optimization iteration process, an update algorithm is introduced, which includes two accepting operators, which are used to update the best and worst working conditions in the current population and the previous generation population, respectively. In response to the fact that the objective function value of the newly generated operating condition is better than the objective function value of the old operating condition, the new operating condition replaces the old operating condition; In response to the fact that the objective function value of the newly generated operating condition is worse than the objective function value of the old operating condition, a decision is made on whether to retain the new operating condition based on the update probability and the discrimination factor. The update algorithm guides the convergence direction of the optimization process to generate a driving condition sequence that meets the target error requirements. Multiple driving condition categories are combined to generate comprehensive driving conditions tailored to different user groups.

2. The method as described in claim 1, characterized in that, The division of the short-stroke segments includes: The segment in the speed-time series from one idle moment to the next idle moment is defined as a short-stroke segment; The short-stroke segment includes an acceleration segment, a deceleration segment, a constant speed segment, and an idle speed segment; The acceleration segment is the driving range in which the acceleration is greater than or equal to a set threshold. The deceleration section is a driving range where the acceleration is less than or equal to a set threshold. The idle speed range is the driving range in which the vehicle speed is less than or equal to a set threshold. The constant speed range is the driving interval where the acceleration is between a set threshold.

3. The method as described in claim 1, characterized in that, The feature parameters include: Average speed, average driving speed, average acceleration during acceleration, average deceleration during deceleration, percentage of acceleration interval, percentage of deceleration interval, percentage of idle speed interval, standard deviation of acceleration, and standard deviation of speed. It also includes the percentage of driving time in multiple speed ranges, used to characterize the driving characteristics of different speed ranges.

4. The method as described in claim 3, characterized in that, The principal component analysis dimensionality reduction method includes: The feature parameter matrix is ​​standardized to eliminate dimensional differences; Principal components are extracted based on the standardized feature parameter matrix, and the cumulative contribution rate of the principal components exceeds a set threshold. The principal components are used to replace the original feature parameters to improve clustering efficiency and working condition construction accuracy.

5. The method as described in claim 4, characterized in that, The cluster analysis method includes: The K-means algorithm is used to cluster the dimensionality-reduced feature parameter matrix; The number of clusters K is determined based on the changing trend of the sum of squared errors within the cluster (SSE). The clustering results are used to classify driving conditions into three categories: low speed, medium speed, and high speed.

6. The method as described in claim 1, characterized in that, The construction of the state transition probability matrix includes: Velocity and acceleration are discretized into multiple states, and state encoding is performed. Based on the state code, the number of transitions between each state is counted to generate a state transition count matrix; The state transition probability is calculated based on the state transition number matrix, and a state transition probability matrix is ​​generated.

7. The method as described in claim 6, characterized in that, The optimization methods of the genetic algorithm include: An initial working condition population is randomly generated based on the state transition probability matrix; The parent load case is selected based on the fitness value of the load case by selecting the operator; State swapping can be performed using the crossover operator without violating the properties of Markov chains; Local state mutations can be performed using mutation operators without violating the properties of Markov chains.

8. The method as described in claim 7, characterized in that, The guided update algorithm includes: In each iteration, the best and worst working conditions are extracted from the current population and the previous generation population, respectively. The optimal working condition is updated and judged. If the objective function value of the new working condition is better than that of the old working condition, it is directly replaced. If the objective function value of the new operating condition is worse than that of the old operating condition, then the decision on whether to retain the new operating condition is based on the update probability and the discrimination factor. The worst-case scenario is updated and judged. If the objective function value of the new scenario is better than that of the old scenario, the old scenario is replaced. If the objective function value of the new operating condition is worse than that of the old operating condition, then the decision on whether to retain the new operating condition is based on the update probability and the discrimination factor. The update probability decreases with the number of iterations to enhance the convergence stability of the algorithm in later stages.

9. A device for constructing vehicle driving conditions based on a guided Markov evolutionary algorithm, characterized in that, include: The data acquisition module is used to collect vehicle driving data and generate standardized speed-time series. The feature extraction module is used to extract multiple feature parameters from the velocity-time series; The feature dimensionality reduction module is used to perform principal component analysis dimensionality reduction processing on the feature parameters; The clustering analysis module is used to classify short-stroke segments based on the dimensionality-reduced feature parameters; The state modeling module is used to construct the state transition probability matrix with two parameters: velocity and acceleration. The genetic algorithm module is used to generate an initial working condition population based on the state transition probability matrix, and to find a better solution based on the initial population; The update algorithm module is used to introduce an update algorithm during the optimization iteration process. It includes two accepting operators, which are used to update the optimal and worst working conditions respectively, and guide the convergence direction of the iterative optimization process. The target control module is used to set the target accuracy of driving conditions and limit the number of iterations in the optimization process, so as to generate a driving condition sequence that meets the target error requirements within the specified number of iterations. The driving condition combination module is used to combine multiple driving condition categories to generate comprehensive driving conditions for the user group.

10. The apparatus as claimed in claim 9, characterized in that, It also includes a velocity-acceleration joint probability density distribution modeling module, used for: Construct the joint probability density distribution matrix of velocity and acceleration; Calculate the similarity coefficient between the operating condition sequence and the original data in the joint velocity-acceleration distribution; The similarity coefficient is used to evaluate the fitting accuracy of the working condition on the joint distribution of velocity and acceleration.

11. The apparatus as claimed in claim 9, characterized in that, It also includes an error control module, used for: The matching degree between the working condition sequence and the original data is calculated based on the multi-feature unified error function; The error function includes the absolute value of the average relative percentage error of multiple characteristic parameters; The error control module is used to determine whether the operating condition meets the target error requirement.

Images