Method for extracting real road emission severe operating condition scenarios based on multivariate statistical analysis

Abstract

The present application relates to a kind of extraction methods of actual road emission severe operating condition scene based on multivariate statistical analysis, belong to engine emission optimization field, including the following steps: S1: by fixed time length 300S time series moving average window method, the signal collected second by second is converted into fixed time series length short trip window data;S2: using factor analysis method, the multiple factor index affecting RDE test is converted into comprehensive trip dynamics factor and terrain factor;S3: using cluster analysis processing short trip data, select severe data segment and analyze its driving cycle and road terrain distribution characteristics, obtain a series of severe data segment instantaneous speed, instantaneous slope change time series diagram;S4: change time series diagram is applied in vehicle chassis dynamometer or hub test stand and carries out RDE test simulation.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of engine emission optimization, and relates to a method for extracting actual road emission severe condition scenes based on multivariate statistical analysis. BACKGROUND

[0002] The internal combustion engine (ICE) of an automobile is the main source of pollution in urban areas, which seriously leads to the deterioration of urban air quality, because according to the data of the United Nations organization, more than 60% of the global population lives in urban areas, which will have a huge adverse effect on the health of urban residents.

[0003] It has become a trend to put forward more stringent regulatory restrictions on road vehicle emissions, but the actual emissions of road traffic are still significantly higher than the expected value, which is due to the fact that the laboratory test cycle cannot well represent the driving emission characteristics of vehicles on actual roads.

[0004] The reason is that the vehicle is generally designed and calibrated using the laboratory standard driving cycle at the present stage. In order to solve these problems, many scholars have developed more targeted driving cycles in combination with regional characteristics to reduce the deviation from the actual road characteristics, which marks an important technical change in the control of road traffic energy consumption and pollutant emissions from laboratory standard test cycle conditions to real road driving scenarios. Since RDE is carried out under real driving conditions, there is no pre-defined driving schedule as in the laboratory driving cycle, so it ensures that acceleration / speed will cover a wider range of vehicle and engine operating conditions, and based on the actual road RDE test data, the related research on severe emission segments is carried out, and a series of working condition segments which can be reproduced on a test bench or a rotating hub are formed, which has practical significance for guiding engine calibration and optimizing engine emission performance.

[0005] The RDE test is introduced in the national sixth emission regulation, and the experimental results will be affected by factors including terrain, road quality, traffic flow, weather, wind speed, temperature and humidity, and the degree of driving behavior aggressiveness, and various reasons lead to long RDE actual test period and high cost;

[0006] Therefore, the companies such as Audi, BMW and Toyota use the European severe condition RTS95 to carry out rotating hub test calibration to increase the RDE test pass rate; but at present, China lacks similar severe calibration conditions, and domestic automobile enterprises are under great development pressure.

[0007] Generally speaking, the current driving cycle construction can be divided into the following two categories:

[0008] The first type is to establish a driving cycle of regional city characteristics based on big data traffic information flow, which mainly considers urban traffic conditions and vehicle driving characteristics, and uses principal component analysis, cluster analysis, short trip division, Markov chain and other methods to extract the driving characteristics of road vehicles and reflect them in the constructed driving cycle.

[0009] The second type is to refer to the construction method of the European severe driving cycle RTS95, which only transforms the original data set into another specific regional vehicle driving data, and the cycle is mainly constructed from the perspective of the driving intensity of the driver RPA during construction, and the core idea is to select the driving cycle with more aggressive driving behavior.

[0010] Among the above two types of technical routes, the first type of driving cycle construction method has been fully researched; the second type of method is also rapidly developing and iterating, and can still be optimized from many angles.

[0011] Most of the above driving cycle construction methods are only constructed from the perspective of vehicle dynamics driving characteristics, and there are few studies on driving cycles based on emission levels and incorporating terrain conditions. At the same time, the European VII emission regulation has more stringent emission limit requirements for RDE tests, and China currently lacks a high dynamic driving cycle RTS95 that can be reproduced on a rotating hub similar to Europe. The driving behavior of this cycle is more aggressive than the WLTC cycle, but it only focuses on the high driving intensity of speed-time data distribution and lacks consideration of terrain conditions and other conditions; Therefore, it is necessary to develop an emission severe driving cycle scenario that meets the road traffic environment in China. Directly selecting the highest emission cycle scenario as the emission severe driving cycle is often affected by the randomness of the sampling data, and does not have typicality and universality, so a severe driving cycle identification and extraction method based on scientific data analysis is of great significance. SUMMARY

[0012] Therefore, the purpose of the present application is to provide a method for extracting actual road emission severe driving cycle scenarios based on multivariate statistical analysis.

[0013] To achieve the above purpose, the present application provides the following technical scheme:

[0014] A method for extracting actual road emission severe driving cycle scenarios based on multivariate statistical analysis, comprising the following steps:

[0015] S1: convert the second-by-second collected signal into a short trip window data with a fixed time length of 300S by a fixed time length 300S time sequence moving average window method;

[0016] S2: use factor analysis method to convert the multiple factor indexes affecting RDE test into two indexes of comprehensive trip dynamics factor and terrain factor;

[0017] S3: using cluster analysis to process short trip data, selecting a representative and representative harsh data segment, and analyzing the driving conditions and road terrain distribution characteristics of the harsh data segment, so as to identify and extract the actual road driving harsh emission scene, and finally obtain a series of instantaneous speed, instantaneous slope change time series of the harsh data segment;

[0018] S4: applying the instantaneous speed, instantaneous slope change time series of the series of harsh data segments to the vehicle chassis dynamometer or hub bench for RDE test simulation.

[0019] Further, in step S1, the signals collected second by second are converted into short trip window data of fixed time length by the fixed time length 300S-moving average window method (300S-MAV), so as to realize filtering and calculate the characteristic parameters of the window and the emission factor.

[0020] Further, in step S2, the window average speed v, the window average acceleration a, the product of the speed and the greater than 0.1 m / s 2 positive acceleration 95th percentile v·a pos

[95] , the 95th percentile value Vsp of the ascending order of the specific power of the vehicle, pos

[95] , the relative positive acceleration R PA , the average positive acceleration M PA , the window cumulative elevation increment CPEI, the method of factor analysis is used to reduce the multi-influence index affecting the RDE test to a comprehensive trip dynamics factor and a terrain factor, so as to convert the multi-dimensional mathematical description of the RDE test boundary condition to a low-dimensional plane system while reducing information loss.

[0021] Further, step S3 specifically includes the following steps:

[0022] S31: using dimensionless processing of different vehicle emission factors, each emission factor is divided by the standard emission factor corresponding to the WLTC operating condition; and in the low-dimensional plane system after the factor analysis, the grid clustering algorithm is used, the data window under the same grid unit is regarded as a data cluster with the same test boundary condition to map the operating condition distribution.

[0023] S32: After dimensionless processing of the window data of different vehicles, the dimensionless emission factors of all window data in the same unit grid are arithmetically averaged and superimposed in the grid clustering, and the obtained value is used to represent the emission level of the region;

[0024] S33: After the grid clustering is completed, the window data in the region with a relatively high overall emission level is selected, and then a second-order clustering is performed to obtain multiple class centroids of the data window, and the data window closest to the centroid in the Euclidean distance is selected as the universal severe emission operating condition scene, and the time sequence changes of the instantaneous vehicle speed and the instantaneous slope of the operating condition scene are obtained.

[0025] The beneficial effects of the present application are that: the present application adopts a fixed time length 300S moving average window method (300S-MAW) to divide the RDE test second-by-second sampling data into a series of data windows, and calculate and obtain the trip attributes, so as to convert the test second-by-second sampling data into fixed time length class trip data objects, which can not only better reflect the local trip characteristics, but also facilitate the application requirements of RDE test operating condition simulation on the engine test bench, form a standard and easy-to-reproduce severe emission scene operating condition segment, and greatly simplify the RDE test procedure from the time angle. Considering the factors affecting the boundary conditions of actual road driving emissions, based on the factor analysis method (extracting main factors by principal component method), the multiple influence indexes affecting the RDE test are reduced and converted into comprehensive trip dynamics factors and terrain factors, so that the multi-dimensional mathematical description of the RDE test boundary conditions is converted to a low-dimensional plane system with less information loss, and the algorithm idea of grid clustering is used on this plane system to map the operating condition distribution by regarding the data windows under the same grid unit as data clusters with the same test boundary conditions. The severe emission scene recognition and extraction method eliminates the influence of abnormal data points and the difference between different vehicle models on the results to a certain extent, and compared with directly selecting the highest emission data segment as the severe emission segment, the severe emission data segment obtained by the method has higher universality. Based on the multivariate statistical analysis of a large amount of RDE test data of multiple light gasoline vehicles, representative centroid data windows are obtained according to road sections and pollutants, and the data windows are arranged according to the application requirements of RDE test operating condition simulation on the engine test bench, a series of universal emission severe segments for testing the vehicle model can be formed, and the severe segments formed can reflect the coupling effect of trip dynamics and driving route terrain on the RDE test. The research can provide a reference for the further improvement of China's emission regulations, guide engine calibration based on actual driving emission characteristics, provide a method reference for the construction of severe driving conditions of road driving emissions, and has important practical significance for urban pollution prevention and reduction of actual road traffic emissions.

[0026] Other advantages, objects, and features of the present application will be apparent to those skilled in the art from the following specification, and will be learned from the study of the following text, or will be taught from the practice of the present application. The objects and other advantages of the present application can be achieved and obtained by the following specification. Attached Figure Description

[0027] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0028] Figure 1 This is a flowchart illustrating the overall process of extracting harsh operating conditions for actual road emissions based on multivariate statistical analysis.

[0029] Figure 2 A schematic diagram of the fixed-duration 300-second moving average window method;

[0030] Figure 3 Dimensionless emission factors of various pollutants for each road segment are shown in the scatter plots. (a) is the scatter plot of dimensionless emission factors of CO2 on the highway, (b) is the scatter plot of dimensionless emission factors of CO on the highway, and (c) is the scatter plot of dimensionless emission factors of NO on the highway. X Dimensionless emission factor dimensionality-reduced scatter plot, (d) is the dimensionless emission factor dimensionality-reduced scatter plot of high-speed PN;

[0031] Figure 4 The images show 3D bar charts and distribution maps of stringent data segments for each road segment, where (a) represents urban CO2 emissions, (b) suburban CO2 emissions, (c) highway CO2 emissions, (d) urban CO2 emissions, (e) suburban CO2 emissions, (f) highway CO2 emissions, and (g) urban NO2 emissions. X Emissions, (h) for suburban NO X Emissions, (i) are high-velocity NO X Emissions, (j) are urban PN emissions, (k) are suburban PN emissions, and (l) are highway PN emissions;

[0032] Figure 5 The following are time-series diagrams showing the typical emission characteristics of various pollutants under severe operating conditions: (a) represents the severe operating condition for CO2 emissions, (b) represents the severe operating condition for CO emissions, and (c) represents the severe operating condition for NO emissions. X Severe emission conditions, (d) is the severe emission conditions for PN. Detailed Implementation

[0033] The present application is illustrated by way of example and not limitation in the figures of the accompanying drawings, in which like references indicate similar, but not necessarily identical, elements. The figures in the following are provided for the purpose of illustration only, and are not intended to limit the present application. In the figures, certain components have been omitted for the sake of clarity. The figures are not drawn to scale and the dimensions are not intended to represent actual dimensions unless explicitly stated.

[0034] The figures are merely schematic and are not intended to limit the present application. Some of the components in the figures can have been exaggerated, omitted, or simplified, and are presented for the purpose of generic illustration only. To better illustrate the present application, some of the components in the figures have been omitted, exaggerated, or simplified, and are not necessarily drawn to scale. It will be apparent to those skilled in the art that certain components in the figures can be omitted, exaggerated, or simplified, and that such omissions, exaggerations, or simplifications are within the scope of the present application.

[0035] The same or similar components in the figures of the present application correspond to the same or similar components. In the description of the present application, it should be understood that the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or positional relationship based on the orientation or positional relationship shown in the figures, and are only used for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore the terms describing the positional relationship in the figures are only used for illustrative purposes, and cannot be understood as limiting the present application. For those skilled in the art, the specific meanings of the above terms can be understood according to the specific circumstances.

[0036] The present application is based on a large number of light-duty gasoline vehicle actual driving emission (RDE) test data for multivariate statistical analysis, and the test second-by-second sampling data is converted into a series of short trip data windows by using a 300S time series moving average window method, and a factor analysis method is used to convert the multiple factor indexes affecting the RDE test into two indexes of comprehensive trip dynamics factor and terrain factor, and then the short trip data is processed by using a clustering analysis, and the representative and representative severe data segments are selected, and the driving cycle and road terrain distribution characteristics are analyzed, so that the identification and extraction of the actual road driving severe emission scene are achieved, and finally a series of instantaneous speed, instantaneous slope change time series diagrams of the severe data segments are obtained, so that the RDE test simulation can be directly applied to the vehicle chassis dynamometer or the hub bench, and the RDE development and calibration of enterprises are supported, so as to meet the urgent needs of the industry for vehicle road actual driving emission control. The work has certain reference value for the further improvement of China's emission regulations, the guidance of engine calibration referring to actual driving emission characteristics, the provision of method reference for the construction of road driving emission severe driving cycle, and the contribution to the urban pollution prevention and reduction of actual road traffic emission. As shown in Figure 1 The present application specifically includes the following steps:

[0037] The data of cold start, parking idling condition is removed from the second-by-second sampling original data. The RDE test continuous second-by-second sampling test data is divided into a series of short trips (data windows) by using a moving average window method (MAW). When dividing the data windows, the moving step forward (or backward) is consistent with the PEMS data sampling period, and the time length of the data window is 300S. The RDE test continuous second-by-second sampling test original data is divided into a series of short trip data windows by using a fixed time length 300S moving average window method (MAW). When dividing the data windows, the moving step forward (or backward) is 1S, which is consistent with the PEMS data sampling period, and the time width of the window is fixed at 300S, until the last window end data is the last processed data. The CO2 data window division schematic diagram is shown in Figure 2 The windows of other pollutants are divided in the same way, and the emission factors of each pollutant in the data window can be calculated.

[0038] The data windows are divided into urban, suburban and highway windows according to the average speed of the data window. Among them, the window with an average speed less than 45km / h is an urban window, the window with an average speed greater than or equal to 45km / h and less than 80km / h is a suburban window, and the window with an average speed greater than or equal to 80km / h and less than 145km / h is a highway window.

[0039] The related parameters representing the attributes of the data window are calculated, including: average speed (v), v·a pos

[95] (car speed and greater than 0.1 m / s 2 95th percentile of positive acceleration product), Vsp pos

[95] (95th percentile value of whole vehicle specific power in ascending order), relative positive acceleration (R PA ), average positive acceleration (M PA ), cumulative positive elevation increment CPEI, etc. The calculated characteristic parameters and pollutant emission factors (specific distance emission) are assigned to the data window as trip attributes, and the data window is given the road segment attribute, so that the test second-by-second sampling data is converted into trip-like data.

[0040] After moving average window processing, a total of 12 data set (12 data sets) of different road segments (urban, suburban, highway), different pollutants (CO2, NO x , CO, PN) are obtained, which are sample data for subsequent statistical analysis.

[0041] In order to select the emission severe segment, principal component analysis is first performed on the window data, and the dimension reduction idea is used to convert multiple indicators into a few comprehensive indicators. When there is a certain correlation between two variables, it can be explained that the information reflected by these two variables has a certain overlap. Principal component analysis is to delete redundant variables (variables with close relationship) from the original variables, and to establish as few new variables as possible, so that these new variables are not correlated with each other, and these new variables maintain the original information as much as possible in reflecting information.

[0042] Because there are characteristic differences between the window data selected from different road segments, the window set is divided into urban, suburban and highway window sets according to the road segment, and factor analysis method is used for processing respectively.

[0043] Factor analysis converts original variables into another set of uncorrelated variables (factors) through coordinate transformation, and then calculates the eigenvalues and corresponding standard orthogonal eigenvectors of the correlation coefficient matrix. According to the eigenvalues of the correlation coefficient matrix, the variance contribution rate and cumulative contribution rate of the component factors are calculated. In the urban window sample set, the cumulative variance contribution rate of the first two factors is 93.4%, which indicates that these two factors basically contain all the information represented by the original variables, and are called component factors. The cumulative variance contribution rate of the first two components in the suburban window sample set is 90.5%, and the cumulative variance contribution rate in the highway window sample set is 91.4%.

[0044] Therefore, in each data window sample set, two component factors composed of five original variables can be used to describe the trip dynamics state of the test vehicle as a whole, so that the test data space is reduced from five dimensions to two dimensions. The information loss in this processing process is small.

[0045] After the factor analysis of each road section, the comprehensive journey dynamic factor is obtained, denoted as FAC1; the terrain feature factor obtained by normalizing the CPEI of the window is used to represent the terrain characteristics of the driving route, denoted as FAC2. FAC1 and FAC2 are regarded as the complete description of the RDE test boundary. Taking the highway section as an example, the dimensionless emission factor scatter plot of the data window after dimension reduction is shown in (a)-(d) of FIG. 6, on the basis of which, the algorithm idea of grid clustering is used to divide a limited number of grid area units on the boundary plane. All data window samples in each grid area unit are regarded as data clusters with similar test boundary conditions, so the arithmetic average of the dimensionless emission factors of all data windows in the same data cluster is added to represent the pollutant emission level under the test boundary. For the data windows of each road section, each pollutant and each vehicle type, the corresponding FAC1 and FAC2 are taken as the horizontal and vertical coordinates, and the arithmetic average of the corresponding emission is taken as the dimensionless emission factor bar coordinate to draw a grid clustering 3D column chart. The dimensionless emission grid clustering 3D column chart of each road section and each pollutant is shown in (a)-(l) of FIG. 7. Grid clustering superimposes data windows of various vehicle types, thereby reflecting the representative characteristics of pollutant emissions of all vehicle types in the grid clustering. From another point of view, it embodies the average value of pollutant emissions of the data window sample group of different vehicle types under the same characteristic test boundary condition, which to some extent represents the emission level of light gasoline vehicles in certain operating condition regions. Figure 3 Figure 4

[0046] According to different road sections (urban, suburban and highway) and different pollutants (CO2, NOx, CO and PN), the grid units with the top 10% of dimensionless emission factors in the grid clustering are selected, and the data windows therein are taken as new research objects, referred to as the simplified data window set; the comprehensive journey dynamic factor FAC1 of each window, the terrain factor FAC2 obtained by normalizing the cumulative positive elevation increment of the window, and the corresponding dimensionless emission factor are taken as inputs for clustering statistics, and the window data with representative (closest to the class centroid) in each class is selected as the typical characteristic representative of the severe operating condition scene of the cluster class. In this paper, the second-order clustering method is used in clustering analysis.

[0047] ​​Two-Step-Cluster (TSC) is different from other traditional clustering methods. It can automatically determine the optimal number of clusters, and deal with the clustering problems of discrete data and continuous data. It is suitable for clustering large data. Two-Step-Cluster is divided into two stages. The first stage is called Pre-Cluster. It calculates the distance of the record data and builds Cluster-Feature-Tree (CF). It generates nodes according to the similarity degree within the node. The second stage uses the clustering results of the first stage to cluster again. Then it uses Akaike-Information-Criterion (AIC) or Bayesian-Information-Criterion (BIC) to judge the clustering results. This paper will use BIC to judge. Finally, the optimal clustering result is obtained. Unlike hierarchical clustering and K-means clustering, Two-Step-Cluster can automatically find the optimal number of clusters under the premise of large data. Two-Step-Cluster has accurate clustering effect and good scalability according to the combination of the two data processing methods.

[0048] The distance is represented by the log-likelihood function as follows:

[0049]

[0050]

[0051] Where: k A represents continuous variables; k B represents discrete variables; L k represents the number of categorical variables; N k represents the number of samples in the kth group; represents the variance of the kth continuous variable; represents the variance of the kth continuous variable in the jth group; N jkl represents the number of the kth categorical variable in the jth group.

[0052] Then the clustering effect is judged by BIC. The calculation method and meaning of BIC are as follows:

[0053]

[0054] Where represents the number of parameters. Since this model uses maximum likelihood method, the maximum value of the likelihood function is represented by .

[0055] According to the results of the second-order clustering, the different working condition scene categories are represented in the form of letters and marked on the grid clustering 3D column chart, so as to facilitate the study of the distribution characteristics of the severe working condition scene on the grid. The grid clustering 3D column chart of each pollutant of each road section and the severe segment labeling are as shown in (a)-(l) of FIG. 15. Figure 4

[0056] For the emission severe data working condition of each pollutant of each road section, the second-by-second sampled data is arranged so as to better describe and reproduce the severe working condition. Since the average vehicle speed, v·a_pos

[95] , Vsp_pos

[95] , R PA , M PA in the characteristic parameters of the severe data segment are closely and directly related to the instantaneous vehicle speed, and the cumulative positive altitude increment in the characteristic parameters directly reflects the instantaneous slope of the road, the instantaneous vehicle speed and the instantaneous slope in the emission severe data working condition are selected to describe the characteristics of the selected emission severe working condition scene. Therefore, the variation of the instantaneous vehicle speed and the instantaneous slope in the emission severe data segment is plotted in the time sequence of 300S, and the time sequence chart of the emission severe working condition with typical characteristics is as shown in (a)-(d) of FIG. 16. Figure 5

[0057] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced equivalently without departing from the purpose and scope of the technical solutions, and all should be covered in the scope of the claims of the present application.​​

Claims

1. A method for extracting real-world road emission severe duty condition scenarios based on multivariate statistical analysis, characterized in that: Comprising the following steps: S1: Convert the second-by-second collected signal into short-trip window data of fixed time length by fixed-length 300S time series moving average window method; S2: Factor analysis is used to transform the multi-factor indicators affecting the RDE experiment into two indicators: a comprehensive travel dynamics factor and a terrain factor; in step S2, the average vehicle speed within the window is comprehensively considered. v Window average acceleration a Vehicle speed greater than 0.1 m / s 2 95th percentile of the product of positive accelerations V·a pos [ 95 The 95th percentile value of the vehicle's power-to-weight ratio in ascending order. Vsp pos [ 95 Relative positive acceleration R PA Mean positive acceleration M PA Cumulative altitude increase within the window CPEI Factor analysis is used to reduce the dimensionality of multiple influencing indicators affecting RDE experiments into comprehensive travel dynamics factors and terrain factors. This reduces information loss and transforms the multidimensional mathematical description of the boundary conditions of RDE experiments into a low-dimensional planar system. On the low-dimensional planar system, a grid clustering algorithm is used to map the working condition distribution by treating data windows under the same grid cell as data clusters with the same experimental boundary conditions. S3: Process the short-trip data using cluster analysis, select the representative and representative severe data segments, and analyze the driving conditions and road terrain distribution characteristics, so as to identify and extract the actual road driving severe emission scene, and finally obtain a series of instantaneous speed, instantaneous slope change time series of severe data segments; S4: Apply the instantaneous speed, instantaneous slope change time series of the series of severe data segments to the whole vehicle chassis dynamometer or hub bench for RDE test simulation.

2. The method for extracting real road emission severe driving condition scenarios based on multivariate statistical analysis according to claim 1, characterized in that: In step S1, the second-by-second collected signal is converted into short-trip window data of fixed time length by fixed-length 300S-moving average window method 300S-MAV, so as to realize filtering, calculation of window characteristic parameters and emission factors.

3. The method of claim 1, wherein: Step S3 specifically comprises the following steps: S31: Use dimensionless processing of different vehicle emission factors, and divide each emission factor by the standard emission factor corresponding to the WLTC condition; S32: After dimensionless processing, the window data of different vehicles are arithmetically averaged and superimposed with the dimensionless emission factors of all window data in the same grid in grid clustering, and the obtained value is used to represent the emission level of the grid; S33: After grid clustering, the window data in the grid unit with the dimensionless emission factor in the top 10% of the grid clustering are selected, and a second-order clustering is performed to obtain multiple class centroids of the data window, and the data window closest to the centroid in the Euclidean distance is selected as the universal severe emission condition scene, and the instantaneous speed and instantaneous slope time series of the condition scene are obtained.

Images