Research On Driving Cycle Of Passenger Car In Northern Environment And Construction Method Of Typical Driving Cycle

Along with the national emission regulation becoming more stringent,passenger car is developing quickly from traditional energy technology towards new energy technology,the power system as an important part of the vehicle technology system needs to adapt to this challenge.At present,the mastery degree for domestic passenger car typical driving characteristics is not sufficient,and the driving cycle that adapts to domestic traffic environment is not generalized widely,which lead to the experiment result in the power system research and development is unable to match the practical level,and cause difficulties in the use of new technology in emission test and energy management.Therefore,developing typical driving cycle that is able to reflect domestic driving characteristics is necessary,which can supply key data support for new energy automobile technology research and development.This paper chose 10 light passenger cars in Changchun city and conducted driving data collection by autonomous driving,established a micro-trip big sample database,the specific application schemes were researched in terms of unsupervised learning clustering,mixed probability distribution model clustering and stochastic process probability model estimation,the typical driving cycles for light passenger car in Changchun city were developed,and the analysis results provide reliable theoretical and applied argument for typical driving construction research.The computation performances of K-means clustering algorithm,density peak clustering algorithm,fuzzy clustering algorithm and self-organizing map neural net clustering algorithm were compared and analyzed by adopting testing sample data set.The results show that,self-organizing map neural net clustering algorithm is able to use a small amount of nerve cells to get the distribution structure of sample set in the feature space,it can establish the corresponding relation between nerve cell and random sample,the clustering for randomsample set is directly completed from the clustering result of nerve cells based on this corresponding relation,this processing manner is conducive to reduce computation burden.This algorithm has the highest clustering precision,the least number of misclassified samples and the highest overlap ratio of position between the clustering central point and the ideal cluster centers,it is suitable for processing the clustering computation of big sample data set.The effect of assembly way of characteristic parameter on clustering computation was researched,the results show that,the data volume of big sample database could be reduced notably and computation efficiency could be enhanced provided that choosing idle proportion,cruising proportion,driving duration,maximum velocity,average driving velocity and velocity standard deviation as characteristic parameters used by describing micro-trip sample,this way can refine medium-low velocity driving classes and avoid class homogenization,and reasonable allocation of samples among classes can be realized.The big sample database was clustered and analyzed by using self-organization map neural net clustering algorithm,the typical driving cycle construction method based unsupervised learning clustering was researched according to the clustering results,medium-low velocity driving cycle and high velocity driving cycle for light passenger car in Changchun city was constructed respectively based on the method.The statistical characteristics were compared between the two constructed driving cycles and respective database,the results show that,the average characteristic parameter deviation is both below3%,the constructed cycles possess high precision and the validity and accuracy of the proposed method is verified.Express way,arterial road in urban district,arterial road in suburb and sub-arterial road were chosen as research object on road grade,the data sets for different road grades were established by extracting sample data from the big sample database,the driving characteristics and the distribution patterns in the feature space for the sample data sets were analyzed statistically.The results show that,there is large overlap area in the feature space for the random sample distribution of arterial road in urban district,arterial road in suburb and sub-arterial road,the three road grades have high similarity of driving pattern.The random samples of express way,arterial road in urban district,arterial road in suburb obey normal distribution.Sub-arterial road has the most abundant driving feature diversity,and the randomsamples of this road grade could not obey strictly normal distribution.Expectation maximization algorithm and discriminant analysis were applied to cluster and analyze the sample data sets of express way and arterial road in urban district.The results show that,the number of samples that the expectation maximization algorithm and discriminant analysis misclassify from the training sample sets is 1 and 0 respectively,the number of fuzzy samples getting from the clustering results of applying the two algorithms to cluster the unknown sample set is 2 and 1 respectively,the results of clustering the rest samples for the two algorithms are the same,the clustering results have high consistency.The typical driving cycle construction method based on clustering algorithm of mixed normal probability distribution model was researched,the clustering results for sample data set getting from Expectation maximization algorithm were applied to construct typical driving cycles for the express road and arterial road in urban district.The compared results show that,the average statistical characteristic parameter deviation between the two constructed driving cycles and respective database is below 3% and 2%,the constructed cycles possess high precision and the validity of the proposed method is verified.The computation method for state transfer probability was discussed based on Markov chain theory.The analysis results show that,maximum likelihood method has deficiency in calculating state transfer probability of small quantity of samples,but modified Kneser-Ney smoothing method is well-adapted to different quantity of samples,it is able to realize the smooth transition for state transfer probability in a certain neighborhood size that the calculating result could conform to driving regularity to a higher degree.The relevance between state transfer probability distribution and normal distribution was examined,the results show that,the shapes of both distributions have highly consistency only on the condition that driving state corresponding to specific speed intervals,in addition to this,both shapes are inconsistent remarkably,unquestioningly applying normal distribution fitting to estimate state transfer probability will bring about more calculating error.The typical driving cycle construction method based on Markov chain stochastic process was researched,typical driving cycles for light passenger car in Changchun city were constructed by use of classified strategy and unclassified strategy according to proposed method.The comparison results show that,the average characteristic parameter deviationbetween the constructed cycles and database are all under 5%,the validity of the proposed constructed method is verified.