CN114048798B - Automobile driving condition construction method based on improved noise reduction self-encoder - Google Patents
Automobile driving condition construction method based on improved noise reduction self-encoder Download PDFInfo
- Publication number
- CN114048798B CN114048798B CN202111227714.2A CN202111227714A CN114048798B CN 114048798 B CN114048798 B CN 114048798B CN 202111227714 A CN202111227714 A CN 202111227714A CN 114048798 B CN114048798 B CN 114048798B
- Authority
- CN
- China
- Prior art keywords
- automobile
- category
- kinematic
- segments
- kinematic segments
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Traffic Control Systems (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention discloses an automobile driving condition construction method based on an improved noise reduction self-encoder, which comprises the following steps: respectively extracting characteristic parameters from the divided kinematic segments, constructing a data set, and constructing a characteristic value matrix by using the characteristic parameters of all the kinematic segments; calculating principal component scores of all the kinematic segments according to the eigenvalue matrix; clustering all the kinematic segments according to the principal component scores; selecting kinematic segments under each category to construct an automobile driving working condition curve, comparing errors between characteristic parameters of each automobile driving working condition curve and characteristic parameters of the data set, and taking the automobile driving working condition curve with the minimum errors as an automobile driving working condition curve to be processed; after the running condition curve of the automobile to be processed is decomposed, the disturbance information is reconstructed with the main information after noise is added, and then the reconstructed information is input into an encoder to obtain an optimized running condition curve of the automobile. The invention increases the accuracy of the running condition of the automobile and is more in line with the actual driving condition.
Description
Technical Field
The invention relates to the field of automobile driving condition construction, in particular to an automobile driving condition construction method based on an improved noise reduction self-encoder.
Background
The driving condition (DRIVING CYCLE) of the automobile is also called a vehicle test cycle, and describes a relation curve of the speed of the vehicle and time under a certain driving environment. As an important basic technology in the automobile industry, the running condition of the automobile plays a vital role in vehicle development, test and evaluation, is the basis of a vehicle energy consumption/emission test method and a limit value standard, and is also a main reference for optimizing various performance indexes of the automobile.
Currently, in vehicle development, test and evaluation, developed countries such as europe and america adopt automobile driving conditions suitable for actual roads in the country. At the beginning of this century, china directly adopts European NEDC running conditions. With the increase of the automobile holding quantity, the gap between the actual oil consumption of the automobile and the legal authentication result is gradually enlarged. Therefore, as the most important basic technology of the automobile industry, it is increasingly important to make the running condition of the automobile which can accord with the actual road of the country. In addition, the Chinese operators are wide, the topography characteristics, the climate conditions, the development conditions and the traffic conditions among cities are also greatly different, and the ideal effect applicable to each city is difficult to achieve by adopting an automobile driving condition. The construction of the driving conditions of the vehicles on actual roads in each city is particularly important and urgent, so that in order to save the cost and improve the reusability, it is necessary to provide a unified method from the collection of data of the original vehicles to the final representative driving conditions of the vehicles.
At present, a common automobile driving condition construction flow is shown in fig. 1, and comprises the following steps:
1) And (5) preprocessing data. The original acquired automobile driving data has a plurality of missing values and abnormal values due to acquisition equipment or environmental factors and the like, and the data which do not meet the conditions are replaced or deleted through operations such as data cleaning, data filling and the like, so that the quality of the data is improved, and the preparation is made for further operation of the data.
2) The kinematic segments are partitioned. And dividing the processed data according to the definition of the kinematic segment. The kinematic segment refers to a driving process of the vehicle between two adjacent parking spots, i.e., a moving process of the vehicle from one idle speed to the next idle speed, and generally includes an idle speed process and a driving process.
3) And extracting characteristic parameters of the kinematic segment to construct a characteristic value matrix. And extracting each kinematic segment according to certain characteristic parameters. The characteristic parameters reflect the driving characteristics of each motion segment, and different researchers or cities can select different characteristic parameters.
4) And carrying out principal component analysis on the eigenvalue matrix. The method adopts a principal component analysis method, namely, a few comprehensive indexes are used for replacing the original multiple indexes, so that the few comprehensive indexes can represent the original multiple indexes as much as possible.
5) And carrying out K-means cluster analysis on the principal component scores. All the kinematic segments are clustered by using a K-means method according to the principal component score, and three categories are generally divided, and the three categories correspond to low-speed, medium-speed and high-speed stages respectively.
6) And constructing the actual running condition of the automobile. And selecting the clustered kinematic segments, and combining the kinematic segments into an automobile driving working condition curve meeting the conditions.
The current construction method for the running condition of the automobile has the following problems:
Firstly, the principal component analysis method presumes that input data meets Gaussian distribution, and actual data is difficult to perfectly meet the Gaussian distribution, so that when the data does not follow the Gaussian distribution, a final result of the principal component analysis can generate deviation, and further a final automobile running condition curve is adversely affected.
Secondly, the K-means clustering algorithm is one of the most widely applied clustering algorithms, but the algorithm has long running time, and particularly, a large amount of data can be collected during the running of the automobile, so that the operation time of the whole clustering process is greatly increased. The K-means algorithm has high requirement on initial point selection, and the initial point selection by people has large uncertainty, so that a final result can generate large errors. In addition, the K-means algorithm is also sensitive to isolated points, so that the quality requirement on the original data is high.
Finally, the actual running condition of the automobile is constructed at present by basically adopting a random search algorithm or a greedy algorithm. The two algorithms are not necessarily capable of finding the globally optimal solution theoretically, and the time complexity is high due to the large search range.
Patent CN110866997A discloses a construction method of running conditions of an electric automobile, which adopts a K-means algorithm based on artificial bee colony to perform cluster analysis on fragments on the basis of the prior art, improves the accuracy and global searching capability of a cluster center, and reduces the accuracy error of a representative working condition curve and actual conditions. However, the scheme has higher time complexity, slower operation speed for a large amount of data, and the constructed running condition curve of the automobile is still inaccurate compared with the actual situation.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: aiming at the technical problems existing in the prior art, the invention provides the automobile driving condition construction method based on the improved noise reduction self-encoder, which increases the accuracy of the automobile driving condition and is more in line with the actual driving condition.
In order to solve the technical problems, the technical scheme provided by the invention is as follows:
A method for constructing an automobile driving condition based on an improved noise reduction self-encoder comprises the following steps:
S1) respectively extracting characteristic parameters of the divided kinematic segments, constructing a data set, and constructing a characteristic value matrix by using the characteristic parameters of all the kinematic segments;
S2) calculating principal component scores of all the kinematic segments according to the eigenvalue matrix;
S3) clustering all the kinematic segments according to the principal component scores to obtain the kinematic segments under different categories respectively;
S4) respectively selecting kinematic segments under each category to construct an automobile running condition curve, comparing errors between characteristic parameters of each automobile running condition curve and characteristic parameters of the data set, and taking the automobile running condition curve with the minimum errors as an automobile running condition curve to be processed;
S5) decomposing the running condition curve of the automobile to be processed to obtain main information and disturbance information, reconstructing the disturbance information with the main information after noise is added to the disturbance information, and inputting the reconstructed information into an automatic encoder to obtain an optimized running condition curve of the automobile.
Further, step S2) includes the steps of:
s21) mapping all samples in the eigenvalue matrix into an eigenvalue space by using a preset nonlinear mapping function to obtain a mapping matrix;
S22) performing principal component analysis on the mapping matrix, extracting a target characteristic value, and calculating a projection value of each kinematic segment on a characteristic vector of the target characteristic value to obtain a principal component score of each kinematic segment.
Further, step S22) includes the steps of:
s221) calculating a kernel matrix by using the mapping matrix and a preset kernel function;
S222) calculating all feature vectors and corresponding feature values of the kernel matrix;
S223) selecting a preset first number of non-zero feature values as target feature values in the order from large to small;
s224) obtaining a mapping value corresponding to each kinematic segment in the mapping matrix, and calculating a projection value of the mapping value corresponding to each kinematic segment on a feature vector corresponding to the target feature value.
Further, step S224) further includes: and normalizing the feature vector corresponding to the target feature value.
Further, step S3) includes the steps of:
S31) taking each kinematic segment as a category;
s32) calculating the similarity between the principal component scores of each category, and respectively attributing two categories with the closest similarity to the same category;
S33) returning to step S32) until all the kinematic segments belong to a preset second number of categories.
Further, step S4) includes the steps of:
s41) screening out the kinematic segments meeting the requirements from each category respectively, and combining all the kinematic segments meeting the requirements in the same category into a candidate set corresponding to the category;
S42) calculating the time occupied by each category, and respectively finding out the combination of the kinematic segments meeting the conditions in the candidate set corresponding to each category according to the time occupied by each category, wherein the combination of the kinematic segments meeting the conditions in the same category forms a set corresponding to the category, and the combination of the kinematic segments is used as an element in the corresponding set;
s43) traversing the sets corresponding to all the categories, respectively selecting elements in the sets corresponding to each category, and synthesizing the elements into an automobile driving working condition curve;
S44) traversing all the automobile running condition curves, comparing errors between the characteristic parameters of the automobile running condition curves and the characteristic parameters of the data set, and taking the automobile running condition curve with the minimum error as the automobile running condition curve to be processed.
Further, in step S41), the step of screening the kinematic segments meeting the requirements from each category includes: for the current category, calculating a clustering center, determining the number n i of the kinematic segments in the candidate set corresponding to the current category according to the total number of the kinematic segments in the current category, calculating the distances from all the kinematic segments in the current category to the clustering center of the current category, and selecting n i kinematic segments according to the sequence from the small distance to the large distance.
Further, in step S42), calculating the time occupied by each category, and finding the combination of the kinematic segments meeting the condition in the candidate set corresponding to each category according to the time occupied by each category specifically includes: and for the current category, calculating the ratio of the total time of the kinematic segments of the current category to the total time of the kinematic segments of all the categories, calculating the product of the ratio and the actual time of the driving working condition of the automobile to obtain the time occupied by the current category, traversing the current category, and selecting the combination of the kinematic segments with the difference between the total time and the time occupied by the current category being smaller than a preset threshold value.
Further, step S5) includes the steps of:
S51) carrying out wavelet decomposition on a running condition curve of the automobile to be processed to obtain main information and disturbance information of each layer, respectively adding Gaussian noise to the disturbance information of each layer, and reconstructing the main information and the disturbance information added with the Gaussian noise;
s52) inputting the reconstructed information into an encoder to obtain a recoded optimized automobile driving condition curve.
Further, the step S1) further includes a step of dividing the kinematic segment, specifically including:
Collecting running data of an automobile in a current city, and replacing or deleting missing values and abnormal values of the running data to obtain processed data;
And acquiring a driving process between two adjacent parking points in the processed data to obtain a set of kinematic segments, wherein the kinematic segments comprise an idling process and a driving process.
Compared with the prior art, the invention has the advantages that:
1. According to the invention, the noise reduction self-encoder is improved, the main information and the disturbance information are obtained by decomposing the automobile running condition curve, the disturbance information is reconstructed with the main information after noise is added, and then the reconstructed information is input into the self-encoder to obtain the optimized automobile running condition curve, so that the error between the characteristic parameters of the automobile running condition curve and the characteristic parameters of the data set can be further reduced, and the representativeness of the automobile running condition is enhanced.
2. According to the invention, the nucleated principal component analysis is adopted, firstly, the nonlinear mapping function is used for mapping the eigenvalue matrix into the characteristic space, so that the mapped data approximately meet Gaussian distribution, and then the principal component analysis is carried out on the mapped data, so that the defect that the traditional principal component analysis can only process the data meeting Gaussian distribution is overcome, the adaptability of the invention is enhanced, and meanwhile, the time complexity is hardly changed.
3. The invention adopts the hierarchical clustering method to cluster the principal component scores of all the kinematic segments, reduces errors caused by manual operation, reduces time complexity, ensures that the method can still ensure faster operation speed for a large amount of data, reduces sensitivity to singular points and isolated points, and enhances the robustness of the method.
4. According to the method, the vehicle running condition curves are constructed based on the candidate set traversal search of the kinematic segments under each category, and under the condition that the time complexity is low, the global optimal solution can be found instead of the local optimal solution, so that the accuracy of the finally constructed vehicle running condition is ensured.
Drawings
FIG. 1 is a flow chart of a conventional method for constructing driving conditions of an automobile.
Fig. 2 is a schematic structural diagram of an improved noise reduction self-encoder according to an embodiment of the present invention.
Fig. 3 is a flowchart of a method for constructing a driving condition of an automobile according to an embodiment of the present invention.
Fig. 4 is a schematic diagram of decomposing a driving condition curve of an automobile to obtain main information and disturbance information in an embodiment of the present invention.
Detailed Description
The invention is further described below in connection with the drawings and the specific preferred embodiments, but the scope of protection of the invention is not limited thereby.
As shown in fig. 2, we propose an improved noise reduction self-encoder based on the noise reduction self-encoder, and the processing procedure is as follows: for the vector x i, firstly, a high-resolution signal and a low-resolution signal are obtained through multi-scale wavelet analysis, gaussian noise is introduced into the high-resolution signal, data of the high-resolution signal are destroyed, the low-resolution signal and the high-resolution signal with increased noise are reconstructed to obtain a new vector x i *, then x i * is input into a self-encoder, the self-encoder reconstructs an input vector x i * according to a preset code z i, noise in the input vector x i * is learned, real input which is not polluted by the noise is obtained, finally, the code z i is obtained, the reconstructed vector x i' is the real input which is not polluted by the noise, and compared with the vector x i, the vector x i more accords with the characteristics of original data.
The improved noise reduction self-encoder can optimize the running condition of the automobile, and the data can be better destroyed by introducing noise from multiple scales, so that the self-encoder automatically learns noise, the robustness and generalization capability of the self-encoder are enhanced, the error of the running condition of the automobile is further reduced, the condition of overfitting is avoided, the final solution is more in accordance with the characteristics of the whole original data, and the representativeness of the running condition of the automobile is enhanced.
As shown in fig. 3, the embodiment provides a method for constructing a driving condition of an automobile based on an improved noise reduction self-encoder, which includes the following steps:
S1) respectively extracting characteristic parameters of the divided kinematic segments, constructing a data set, and constructing a characteristic value matrix by using the characteristic parameters of all the kinematic segments;
S2) calculating principal component scores of all the kinematic segments according to the eigenvalue matrix;
S3) clustering all the kinematic segments according to the principal component scores to obtain the kinematic segments under different categories respectively;
S4) respectively selecting kinematic segments under each category to construct an automobile running condition curve, comparing errors between characteristic parameters of each automobile running condition curve and characteristic parameters of the data set, and taking the automobile running condition curve with the minimum errors as an automobile running condition curve to be processed;
S5) decomposing the running condition curve of the automobile to be processed to obtain main information and disturbance information, reconstructing the disturbance information with the main information after noise is added to the disturbance information, and inputting the reconstructed information into an automatic encoder to obtain an optimized running condition curve of the automobile.
The step S1) is preceded by the step of dividing the kinematic segment, and specifically comprises the following steps:
Collecting running data of an automobile in a current city, replacing or deleting missing values and abnormal values of the running data, and generally processing the missing values and the abnormal values in the running data through operations such as data cleaning and data filling to obtain processed data;
And acquiring a driving process between two adjacent parking points in the processed data to obtain a set of kinematic segments, wherein the kinematic segments comprise an idling process and a driving process.
In this embodiment, step S1) selects 14 feature parameters: the calculation formulas of the average speed v m, the average running speed v mr, the maximum speed v max, the vehicle speed standard deviation v sd, the (acceleration segment) average acceleration a am, the (deceleration) average acceleration a dm, the maximum acceleration a max, the minimum acceleration (maximum deceleration) a min, the acceleration standard deviation a asd, the deceleration standard deviation a dsd, the acceleration time ratio P a, the deceleration time ratio P d, the idle time ratio P i and the uniform speed time ratio P u, and the 14 characteristic parameters are respectively as follows:
a) Average speed v m, in km/h, defined as the arithmetic mean of the vehicle speed over a period of time:
wherein T is total time, N is the time length of the total time T, and v i is the speed of the vehicle at the moment i.
B) The average running speed v mr, the unit is km/h, is defined as the arithmetic average of the speed of the automobile in a running state in a certain time:
wherein T i is the time when the vehicle is in idle state in the period of time.
C) Maximum speed v max, in km/h, defined as the maximum value of the vehicle speed in a certain time:
vmax＝max{vi,i＝1,2,…,N} (3)
d) The standard deviation v sd of the vehicle speed, the unit is km/h, is defined as the standard deviation of all vehicle speeds in a certain time:
e) Acceleration section average acceleration a am, in m/s 2, is defined as the arithmetic average of the acceleration of the vehicle during acceleration over time:
Wherein, For acceleration at time i in the acceleration process, T a is the total time of the acceleration process;
The calculation formula of the acceleration is as follows:
Where a i,i+1 is the acceleration from time i to time i+1, and t i is the time length i.
F) The average acceleration a dm of the deceleration section, the unit is m/s 2, is defined as the arithmetic average value of the acceleration of the vehicle during deceleration in a certain period of time:
Wherein, To the acceleration at time i during deceleration, T d is the total time of the deceleration process.
G) Maximum acceleration a max, in m/s 2, is defined as the maximum value of vehicle acceleration over time:
amax＝max{ai,i＝1,2,…,N-1} (8)
Wherein a i is the acceleration at time i.
H) Minimum acceleration (maximum deceleration) a min is in m/s 2. Defined as the minimum value of vehicle acceleration over time:
amin＝min{ai,i＝1,2,…,N-1} (9)
Wherein a i is the acceleration at time i.
I) The standard deviation a asd of acceleration, the unit is m/s 2, is defined as the standard deviation of acceleration of the vehicle in the acceleration process in a certain time period:
Wherein N 1 is the total time length of the acceleration process in the specified time, For the acceleration at time i in the acceleration process, a am is the average acceleration of the acceleration section.
J) The standard deviation a dsd of the deceleration, the unit is m/s 2, is defined as the standard deviation of the acceleration speed of the vehicle in the deceleration process in a certain time period:
Wherein N 2 is the total time length of the deceleration process in the appointed time, For the acceleration at time i in the deceleration process, a dm is the average acceleration of the deceleration section.
K) The acceleration time ratio P a is defined as the percentage of the total time that the vehicle is in an acceleration state within a certain period:
wherein T a is the total time of the acceleration process, and T is the total time.
L) the deceleration time ratio P d, defined as the percentage of the total time that the vehicle is in a decelerating state for a certain period of time:
Wherein T d is the total time of the deceleration process, and T is the total time.
M) the idle time ratio P i, defined as the percentage of the total time that the vehicle is in idle state for a certain period:
wherein T i is the time of the vehicle in the idle state in the period of time, and T is the total time.
N) the constant speed time ratio P u is defined as the percentage of the total time of the vehicle in a constant speed state within a certain time:
Pu＝1-Pa-Pd-Pi (14)
Wherein, P a is the acceleration time duty ratio, P d is the deceleration time duty ratio, and P u is the uniform velocity time duty ratio.
Step S2) of the embodiment adopts KERNEL PCA based on hyperbolic tangent function to perform dimension reduction processing on the characteristic parameters in the characteristic value matrix, compared with PCA, KERNEL PCA can extract nonlinear characteristics of data more effectively under the condition that the original data distribution condition has no requirement, overcomes the defect that the traditional PCA can only process data meeting Gaussian distribution, enhances the adaptability of an algorithm, and has little increase in the time complexity of the algorithm, and the method comprises the following steps:
s21) mapping all samples in the eigenvalue matrix into an eigenvalue space F by using a preset nonlinear mapping function phi to obtain a mapping matrix, so that data in the mapping matrix approximately meets Gaussian distribution;
S22) performing principal component analysis on the mapping matrix, extracting a target feature value, and calculating a projection value of each kinematic segment on a feature vector of the target feature value to obtain a principal component score of each kinematic segment, wherein the principal component score comprises the following steps:
S221) calculating a kernel matrix K by using the mapping matrix and a preset kernel function;
In this embodiment, the kernel function is a hyperbolic tangent function (Hyperbolic tangent), the expression of which is shown in formula (15), and the calculation of the corresponding kernel matrix K is shown in formula (16):
K(x,y)＝tanh(g(x·y)+θ) (15)
Ki,j＝(Φ(xi)·Φ(xj))＝K(xi,xj) (16)
Where i, j=1, 2, …, N, K is an nxn matrix, Φ (x i) is a mapping value of x i, Φ (x j) is a mapping value of x j, and x i、xj represents the eigenvalues of the ith and jth kinematic segments, respectively.
S222) calculating all eigenvectors and corresponding eigenvalues of the kernel matrix K;
s223) selecting the first M non-zero eigenvalues as the target eigenvalues in the order from large to small, and normalizing the eigenvectors corresponding to the target eigenvalues, where the formula is as follows:
wherein lambda m represents the mth non-zero eigenvalue, The m non-zero eigenvalue of the ith and jth kinematic segment is represented, alpha m represents the corresponding eigenvector of the m non-zero eigenvalue, K i,j is an element in the kernel matrix, i, j=1, 2, …, N, N eigenvectors are arranged as alpha 1,α2,…,αm,…,αN according to the size of the corresponding eigenvalue;
S224) obtaining a mapping value corresponding to each kinematic segment in the mapping matrix, and respectively calculating a projection value of the mapping value corresponding to each kinematic segment on a feature vector corresponding to the target feature value according to the formula (18), so as to obtain a principal component score of each kinematic segment.
Wherein phi (x i) is the mapping value of x i, phi (x) is the mapping value of test data x,The mth non-zero eigenvalue of the ith kinematic segment is represented, and x i represents the eigenvalue of the ith kinematic segment.
Step S3) of the embodiment adopts BIRCH hierarchical clustering to cluster all the kinematic segments according to the principal component score, the BIRCH hierarchical clustering is totally called as balanced iteration protocol and clustering (Balanced Iterative Reducing and Clustering Using Hierarchies, hereinafter abbreviated as BIRCH hierarchical clustering) using a hierarchical method, errors caused by manual operation are reduced, the time complexity of an algorithm is reduced, the method can still ensure a faster running speed for a large amount of data, meanwhile, the sensitivity to singular points and isolated points is reduced, the robustness of the method is enhanced, and the step S3) comprises the following steps:
S31) taking each kinematic segment as a category;
S32) calculating the similarity between principal component scores of each category, and assigning two categories closest to each other as the same category, respectively, including:
Selecting a category a from the set A of all categories as a reference category, and calculating the similarity between principal component scores of other categories except the reference category and the reference category respectively;
The reference class (namely class a) and the class b with the highest similarity with the reference class are attributed to a new class c, wherein the new class c contains elements in the class a and the class b, and the class a and the class b are removed from the set;
repeating the steps until the set A is empty, and forming a new set A1 by all new categories;
S33) returns to step S32) until all the kinematic segments belong to 3 categories, which correspond to low, medium and high speeds, respectively.
Step S4) of the present embodiment, a candidate set-based traversal search method is adopted to build an automobile driving condition curve, filtering is performed according to errors between the characteristic parameters of each curve and the characteristic parameters of the whole data set in step S1), and under the condition that the time complexity is not high, a global optimal solution rather than a local optimal solution can be found, and step S4) includes the following steps:
s41) respectively screening out the kinematic segments meeting the requirements from each category, combining all the kinematic segments meeting the requirements in the same category into a candidate set corresponding to the category, filtering a plurality of kinematic segments with weak correlation with the data set in the step S1) by the candidate set, reducing the time required by subsequent traversal search, and screening out the kinematic segments meeting the requirements specifically comprises the following steps: for the current category, calculating a clustering center, and determining the number n i,ni of the kinematic segments in the candidate set corresponding to the current category according to the total number of the kinematic segments in the current category, wherein the expression is as follows:
ni＝Ni//3 (19)
Wherein N i represents the total number of kinematic segments of the current class, i=1, 2,3, the sequence number of the current class,// symbol represents integer division;
Then calculating the distances from all the kinematic segments in the current category to the clustering center of the current category, and selecting n i kinematic segments according to the sequence from small to large, wherein the kinematic segments are the kinematic segments meeting the requirements;
S42) calculating the time occupied by each category, and respectively finding out the combination of the kinematic segments meeting the conditions in the candidate set corresponding to each category according to the time occupied by each category, wherein the combination of the kinematic segments meeting the conditions in the same category forms a set corresponding to the category, and the combination of the kinematic segments is used as an element in the corresponding set; further, calculating the time occupied by each category, and finding out the combination of the kinematic segments meeting the conditions in the candidate set corresponding to each category according to the time occupied by each category specifically includes: for the current category, calculating the ratio of the total time of all the kinematic segments of the current category to the total time of all the kinematic segments of all the categories, and calculating the product of the ratio and the actual time of the driving working condition of the automobile to obtain the time occupied by the current category, namely calculating the time occupied by the current category according to a formula (20):
Wherein T is total time of all the kinematic segments of the current category, T i is total time of the kinematic segments of the current category, i=1, 2,3, a serial number of the current category, and T is actual time of the running condition of the automobile, which is 1800 seconds in the embodiment;
then traversing the current category, and selecting a combination of the kinematic segments, wherein the difference between the total time and the time occupied by the current category is smaller than a preset threshold, and the threshold in the embodiment is +/-5%, for example, the difference between the total time of the combination of the kinematic segment a and the kinematic segment b in the current category and the time occupied by the current category is smaller than the preset threshold, and the difference between the total time of the combination of the kinematic segment b and the kinematic segment c and the time occupied by the current category is also smaller than the preset threshold, so that the combination of the kinematic segment a and the kinematic segment b, and the combination of the kinematic segment b and the kinematic segment c are all combinations of the kinematic segments meeting the condition;
In this embodiment, the combinations of the kinematic segments satisfying the conditions of the same category form the sets G 1、G2、G3 corresponding to the low speed, the medium speed, and the high speed, respectively;
s43) traversing the sets corresponding to all the categories, respectively selecting elements in the sets corresponding to each category, and synthesizing the elements into an automobile driving working condition curve, wherein the method specifically comprises the following steps:
S431) selecting an element in the set G 1;
S432) selecting an element in the set G 2;
S433) selecting elements in the set G 3, and synthesizing the elements selected in the set G 1、G2、G3 into an automobile driving condition curve;
S434) returning to step S433) until the elements in the set G 3 are selected;
S435) returning to the step S432) until the elements in the set G 2 are selected;
s436) returning to step S431) until the elements in the set G 1 are selected;
S44) traversing all the automobile running condition curves, comparing errors between the characteristic parameters of the automobile running condition curves and the characteristic parameters of the data set, and taking the automobile running condition curve with the minimum error as the automobile running condition curve to be processed.
Step S5) of the embodiment optimizes the running condition curve of the automobile by using the improved noise reduction self-encoder, so that the error of the running condition of the automobile can be further reduced, the final solution is more in accordance with the characteristics of the whole original data, the representativeness of the running condition of the automobile is enhanced, and the step S5) comprises the following steps:
S51) carrying out wavelet decomposition on a running condition curve of the automobile to be processed to obtain main information and disturbance information of each layer, respectively adding Gaussian noise to the disturbance information of each layer, and reconstructing the main information and the disturbance information added with the Gaussian noise;
s52) inputting the reconstructed information into an encoder to obtain a recoded optimized automobile driving condition curve.
As shown in fig. 4, in step S51) of the present embodiment, three-layer wavelet decomposition is adopted for the vehicle running condition curve to be processed, the low-resolution information V 3 finally obtained contains main information of data, the high-resolution information W 1、W2、W3 contains disturbance information, the gaussian noise is increased to W 1、W2、W3 by setting a threshold value, then the low-resolution information V 3 and the processed high-resolution information W 1、W2、W3 are reconstructed, and the reconstructed data are input into the self-encoder, so that the recoded optimized vehicle running condition curve can be obtained, and according to the effects described above, the vehicle running condition curve is optimized, compared with the vehicle running condition curve to be processed, the error between the characteristic parameters and the characteristic parameters of the total dataset is further reduced, and the representativeness of the vehicle running condition is enhanced.
Compared with the conventional method for constructing the running working conditions of the automobile, the method of the embodiment can greatly improve the representativeness of the running working condition curve of the automobile, increase the accuracy of the running working conditions of the automobile, reduce the error between the running working condition curve and the original data set, and more accords with the actual driving condition, which is also the key point of the improvement of the method of the embodiment; meanwhile, the adaptability of the method is more general, the requirement on the original data is lower, and the data extracted from the original data is more true; in addition, the method of the embodiment has lower time complexity, lower calculation cost and stronger calculation capability, has stronger processing capability on a large amount of data and accords with the current times of trend of big data.
The foregoing is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. While the invention has been described with reference to preferred embodiments, it is not intended to be limiting. Therefore, any simple modification, equivalent variation and modification of the above embodiments according to the technical substance of the present invention shall fall within the scope of the technical solution of the present invention.
Claims (6)
1. The automobile driving condition construction method based on the improved noise reduction self-encoder is characterized by comprising the following steps of:
S1) respectively extracting characteristic parameters of the divided kinematic segments, constructing a data set, and constructing a characteristic value matrix by using the characteristic parameters of all the kinematic segments;
s2) calculating principal component scores of all the kinematic segments according to the eigenvalue matrix;
S3) clustering all the kinematic segments according to the principal component scores to obtain the kinematic segments under different categories respectively;
S4) respectively selecting kinematic segments under each category to construct an automobile running working condition curve, comparing errors between characteristic parameters of each automobile running working condition curve and characteristic parameters of the data set, and taking the automobile running working condition curve with the minimum errors as an automobile running working condition curve to be processed, wherein the method comprises the following steps of:
S41) respectively screening out the kinematic segments meeting the requirements from each category, combining all the kinematic segments meeting the requirements in the same category into a candidate set corresponding to the category, and respectively screening out the kinematic segments meeting the requirements from each category, wherein the step of respectively screening out the kinematic segments meeting the requirements specifically comprises the following steps: for the current category, calculating a clustering center, and determining the number of the candidate set kinematic segments corresponding to the current category according to the total number of the kinematic segments in the current category Calculating the distances from all the kinematic segments in the current category to the clustering center of the current category, and selecting according to the sequence from small distances to large distancesA kinematic segment;
S42) calculating the time occupied by each category, finding out the combination of the kinematic segments meeting the conditions in the candidate set corresponding to each category according to the time occupied by each category, wherein the combination of the kinematic segments meeting the conditions in the same category forms a set corresponding to the category, the combination of the kinematic segments is taken as an element in the corresponding set, the time occupied by each category is calculated, and the combination of the kinematic segments meeting the conditions is found out in the candidate set corresponding to each category according to the time occupied by each category, wherein the method specifically comprises the following steps: for the current category, calculating the ratio of the total time of the kinematic segments of the current category to the total time of the kinematic segments of all the categories, calculating the product of the ratio and the actual time of the driving working condition of the automobile to obtain the time occupied by the current category, traversing the current category, and selecting the combination of the kinematic segments with the difference between the total time and the time occupied by the current category being smaller than a preset threshold value;
s43) traversing the sets corresponding to all the categories, respectively selecting elements in the sets corresponding to each category, and synthesizing the elements into an automobile driving working condition curve;
s44) traversing all the automobile running condition curves, comparing errors between characteristic parameters of the automobile running condition curves and characteristic parameters of a data set, and taking the automobile running condition curve with the minimum error as an automobile running condition curve to be processed;
S5) decomposing the running condition curve of the automobile to be processed to obtain main information and disturbance information, reconstructing the disturbance information with the main information after noise is added to the disturbance information, and inputting the reconstructed information into an automatic encoder to obtain an optimized running condition curve of the automobile, wherein the method comprises the following steps of:
s51) carrying out wavelet decomposition on a running condition curve of the automobile to be processed to obtain main information and disturbance information of each layer, respectively adding Gaussian noise to the disturbance information of each layer, and reconstructing the main information and the disturbance information added with the Gaussian noise;
S52) inputting the reconstructed information into an encoder to obtain a recoded optimized automobile driving condition curve.
2. The method for constructing the running condition of the automobile based on the improved noise reduction self-encoder as claimed in claim 1, wherein the step S2) comprises the steps of:
S21) mapping all samples in the eigenvalue matrix into an eigenvalue space by using a preset nonlinear mapping function to obtain a mapping matrix;
s22) performing principal component analysis on the mapping matrix, extracting a target characteristic value, and calculating a projection value of each kinematic segment on a characteristic vector of the target characteristic value to obtain a principal component score of each kinematic segment.
3. The method for constructing a driving condition of an automobile based on an improved noise reduction self-encoder according to claim 2, wherein the step S22) comprises the steps of:
s221) calculating a kernel matrix by using the mapping matrix and a preset kernel function;
S222) calculating all feature vectors and corresponding feature values of the kernel matrix;
s223) selecting a preset first number of non-zero feature values as target feature values in the order from large to small;
S224) obtaining a mapping value corresponding to each kinematic segment in the mapping matrix, and calculating a projection value of the mapping value corresponding to each kinematic segment on a feature vector corresponding to the target feature value.
4. The method for constructing a driving condition of an automobile based on an improved noise reduction self-encoder as claimed in claim 3, wherein the step S224) further comprises: and normalizing the feature vector corresponding to the target feature value.
5. The method for constructing the running condition of the automobile based on the improved noise reduction self-encoder as claimed in claim 1, wherein the step S3) comprises the steps of:
s31) taking each kinematic segment as a category;
S32) calculating the similarity between the principal component scores of each category, and respectively attributing two categories with the closest similarity to the same category;
S33) returning to step S32) until all the kinematic segments belong to a preset second number of categories.
6. The method for constructing the running condition of the automobile based on the improved noise reduction self-encoder according to any one of claims 1 to 5, wherein the step S1) further comprises a step of dividing the kinematic segment, specifically comprising:
Collecting running data of an automobile in a current city, and replacing or deleting missing values and abnormal values of the running data to obtain processed data;
And acquiring a driving process between two adjacent parking points in the processed data to obtain a set of kinematic segments, wherein the kinematic segments comprise an idling process and a driving process.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111227714.2A CN114048798B (en) | 2021-10-21 | 2021-10-21 | Automobile driving condition construction method based on improved noise reduction self-encoder |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111227714.2A CN114048798B (en) | 2021-10-21 | 2021-10-21 | Automobile driving condition construction method based on improved noise reduction self-encoder |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114048798A CN114048798A (en) | 2022-02-15 |
| CN114048798B true CN114048798B (en) | 2024-08-20 |
Family
ID=80205489
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111227714.2A Active CN114048798B (en) | 2021-10-21 | 2021-10-21 | Automobile driving condition construction method based on improved noise reduction self-encoder |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114048798B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114676211B (en) * | 2022-03-08 | 2025-08-05 | 广州市香港科大霍英东研究院 | Method, device, equipment and storage medium for generating vehicle driving data |
| CN115099365B (en) * | 2022-07-22 | 2025-03-14 | 云南中烟工业有限责任公司 | A method for noise reduction, dimension reduction and feature extraction of production section data |
| CN115420507B (en) * | 2022-08-17 | 2023-07-11 | 武汉理工大学 | Method for establishing engine fuel economy bench test working condition |
| CN115979350A (en) * | 2023-03-20 | 2023-04-18 | 北京航天华腾科技有限公司 | Data acquisition system of ocean monitoring equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2013105869A1 (en) * | 2012-01-13 | 2013-07-18 | Pulse Function F6 Limited | Telematics system with 3d inertial sensors |
| CN113052196A (en) * | 2019-12-27 | 2021-06-29 | 南京理工大学 | Automobile working condition construction method based on greedy algorithm |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109409202B (en) * | 2018-09-06 | 2022-06-24 | 惠州市德赛西威汽车电子股份有限公司 | Robust lane line detection method based on dynamic interesting area |
| CN109932191B (en) * | 2019-03-15 | 2021-10-15 | 中国汽车技术研究中心有限公司 | An actual road driving condition library and construction method |
-
2021
- 2021-10-21 CN CN202111227714.2A patent/CN114048798B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2013105869A1 (en) * | 2012-01-13 | 2013-07-18 | Pulse Function F6 Limited | Telematics system with 3d inertial sensors |
| CN113052196A (en) * | 2019-12-27 | 2021-06-29 | 南京理工大学 | Automobile working condition construction method based on greedy algorithm |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114048798A (en) | 2022-02-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114048798B (en) | Automobile driving condition construction method based on improved noise reduction self-encoder | |
| CN114708434B (en) | Cross-domain remote sensing image semantic segmentation method based on iterative intra-domain adaptation and self-training | |
| CN114005085B (en) | Method for detecting and counting distribution of dense crowd in video | |
| CN114120102A (en) | Boundary-optimized remote sensing image semantic segmentation method, device, equipment and medium | |
| CN102073748B (en) | Visual keyword based remote sensing image semantic searching method | |
| CN113297795A (en) | Method for constructing running condition of pure electric vehicle | |
| CN112950615B (en) | Thyroid nodule invasiveness prediction method based on deep learning segmentation network | |
| CN109871875A (en) | A method of building change detection based on deep learning | |
| CN114139618B (en) | Signal-dependent noise parameter estimation method based on improved density peak clustering | |
| CN111104398A (en) | Detection method and elimination method for approximate repeated record of intelligent ship | |
| CN113034471B (en) | SAR image change detection method based on FINCH clustering | |
| CN108388656A (en) | A kind of image searching method based on label correlation | |
| CN113205107A (en) | Vehicle type recognition method based on improved high-efficiency network | |
| CN113962600A (en) | Traffic safety risk diagnosis method and device | |
| CN116612021B (en) | Image redundancy removal method based on self-supervised ranking learning | |
| CN118469969A (en) | Organic pore identification and pore parameter determination method of shale and medium | |
| CN116664899B (en) | Remote sensing image road extraction method based on fuzzy C-means clustering | |
| CN110442736A (en) | A Semantic Enhanced Subspace Cross-media Retrieval Method Based on Quadratic Discriminant Analysis | |
| CN114169247B (en) | Method, device, equipment and computer-readable storage medium for generating simulated traffic flow | |
| CN118990124B (en) | A tool wear state monitoring method based on DDPM diffusion model | |
| CN117036373B (en) | Large-scale point cloud segmentation method based on round function bilinear regular coding and decoding network | |
| CN117708583A (en) | Rough set attribute reduction method | |
| CN115718886A (en) | Method and device for constructing driving conditions of pure electric vehicles | |
| CN116166708B (en) | An operation platform management system based on big data analytics | |
| CN116721302B (en) | Ice and snow crystal particle image classification method based on lightweight network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |