CN115798239A - Vehicle operation road area type identification method - Google Patents

Abstract

The invention discloses a vehicle operation road area type identification method, which comprises the following steps: the method comprises the following steps: dividing a mountain area type, an urban area type, a suburban area type and a high-speed area type in a target area; and the range of the area type is embodied as discrete longitude and latitude point information; step two: judging the area type of each short journey according to the vehicle GPS positioning information and the area type range information; step three: adding labels of the types of the areas to which the vehicles run in each short-stroke characteristic parameter database; and meanwhile, a machine learning database is obtained, a machine learning algorithm is utilized for training, and a vehicle operation road region classification model based on machine learning is constructed and is used for identifying a vehicle operation road region. The method and the device realize acquisition of the region type label of the user data before synthesis of the driving conditions, can be used for constructing the driving conditions of different city partition types, and can calibrate technical parameters timely and accurately to evaluate the performance of the whole vehicle.

Description

A type identification method of vehicle running road area

technical field

The invention relates to the technical field of vehicle running road area type identification, in particular to a vehicle running road area type identification method.

Background technique

The global energy shortage and environmental pollution problems are becoming more and more serious. As one of the necessary tools for people to travel, the energy consumption and emission of automobiles have a decisive impact on the overall situation. Therefore, in order to meet the increasingly stringent emission regulations of various countries The strategy of friendly and sustainable development requires that all automobile manufacturers are paying more attention to the fuel economy of fuel vehicles and the energy consumption of electric vehicles.

Driving condition is a curve describing the speed and time of vehicle driving characteristics, which reflects the most representative vehicle driving characteristics of a certain model in a certain area or a certain road type. It is mainly used to determine vehicle pollutant emissions and Fuel consumption is an important basis for vehicle energy consumption and emission testing. It can also be used to analyze the damage and life of the vehicle's drive motor, providing a reference for the technical development and evaluation of new models. Previously, most of the automotive test conditions used in the industry were international or domestic standard conditions, but in fact, the social, economic and geographical characteristics of each city are not the same, and even have great differences. Therefore, typical standard operating conditions are not sufficient to represent the actual road driving conditions in different regions or cities. Not only that, but the road traffic conditions in different geographical areas of the same city are very different. Simple accumulation is not conducive to obtaining typical working conditions with strong adaptability, and cannot more accurately reflect the driving characteristics of specific research objects. In order to make the constructed driving conditions more detailed and representative, it is a relatively novel solution to construct driving conditions of different city subregions based on geographic location information.

The driving conditions of sub-areas are the basis for powertrain parameter matching and control strategy optimization in vehicle R&D and design. This is because the driving conditions of different types of regions have different characteristics, and the optimal control parameters and forms of vehicles are also suitable. different.

In order to be able to construct the driving conditions of typical urban subregions, and to adjust the vehicle control parameters in real time according to the actual driving conditions of the vehicle, and to calibrate the technical parameters in a timely and accurate manner to evaluate the performance of the vehicle, a set of methods is required. Identify the type of area in which the vehicle operates.

Contents of the invention

The purpose of the present invention is to provide a vehicle operation road area type identification method, which can not only be used for the preliminary preparation of the development of typical working conditions, but also can be used for real-time working condition identification, improve vehicle operation conditions, and serve big data analysis for vehicle operation cloud Provide data information.

In order to achieve the above object, the technical scheme provided by the invention is as follows:

A method for identifying a vehicle running road area type, characterized in that it includes:

Step 1: According to the geographical location information, divide the mountain area type in the target area; according to the road type, divide the target area into urban area type, suburban area type and high-speed area type; and represent the range of the area type as discrete latitude and longitude points status information;

Step 2: The user data is used as input. The user data includes vehicle operation data information and vehicle GPS positioning information. After preprocessing the data, divide the original data into several kinematic segments according to the short-stroke method, and calculate The characteristic parameters of each short trip are obtained from the short trip characteristic parameter database; according to the vehicle GPS positioning information and the area type range information, the area type to which each short trip belongs is judged;

Step 3: In each short-travel characteristic parameter database, add the label of the area type of vehicle operation; at the same time, obtain the machine learning database, use the machine learning algorithm for training, and build a vehicle operation road area classification model based on machine learning. This model will use It is used to identify the road area where the vehicle runs.

Optionally, the first step specifically includes:

According to the scope of urban administrative divisions, determine the boundary line between the urban area and the suburban area; divide the boundary between the city and the suburbs into the urban area, and obtain the GPS point information of the urban area boundary;

All expressways in the urban administrative area are classified as high-speed areas, and GPS point information of high-speed areas is obtained;

According to the DEM elevation data, calculate the terrain relief, carry out neighborhood analysis, reclassify the terrain relief, obtain the range of mountainous areas, and obtain the GPS point information of the boundary of the mountainous area;

Areas other than urban areas, high-speed areas, and mountainous areas are classified as suburban areas, and GPS point information on the borders of suburban areas is obtained.

Optionally, the boundary line between the urban area and the suburban area is the Ring Expressway or the Ring Line.

Optionally, the second step specifically includes:

The user's short-stroke data is classified according to the type of area; after preprocessing the original data including vehicle operation data information and vehicle GPS positioning information, the short-stroke is divided according to the short-stroke method, and the original data is divided into several kinematic segments, and the user Short-stroke database, calculating the characteristic parameters of each short-stroke for subsequent use;

Construct a function to identify the type of area to which the short trip belongs. The function takes the GPS positioning information of the short trip and the GPS information of the four types of areas as input, and outputs the label of the area type to which the short trip belongs.

Optionally, the step three specifically includes:

Use the random forest algorithm of machine learning to build a classification model of the vehicle running road area, and obtain the machine learning database, including user short-distance travel data, characteristic parameters and labels, use the random forest algorithm to train the database, obtain the random forest model, and adjust the machine learning reasonably The data in the database is used as the ratio of the training set and the test set. According to the realization effect, the model parameters are adjusted, and finally a machine learning-based vehicle operating road area classification model is constructed.

Optionally, the characteristic parameters include:

Running time, running distance, running speed, acceleration time, deceleration time, constant speed time, idle time, acceleration ratio, deceleration ratio, uniform speed ratio, idle speed ratio, maximum speed, average speed, speed standard deviation, maximum acceleration, average acceleration in acceleration section , standard deviation of positive acceleration, maximum deceleration, average deceleration of deceleration section, standard deviation of negative deceleration, average acceleration, standard deviation of acceleration, relative positive acceleration, ratio of different speed intervals, number of accelerations, number of decelerations, number of constant speeds, number of stops , standard deviation of torque, average positive torque, average negative torque, maximum positive torque, maximum negative torque, average torque of idle speed segment, average torque of running segment, average positive torque of running segment, average negative rotation of running segment Torque, positive torque time ratio, negative torque time ratio, maximum fluctuation when torque increases, maximum fluctuation when torque decreases, standard deviation of torque fluctuation, average fluctuation when torque increases, average fluctuation when torque decreases amount, time ratio of high torque interval, time ratio of medium torque interval, time ratio of low torque interval, absolute maximum value of yaw rate, absolute maximum value of lateral acceleration, braking times, duration of acceleration segment, energy consumption per 100 kilometers, Motor total damage, front motor total damage, rear motor total damage, motor unit mileage damage, front motor unit mileage damage, rear motor unit mileage damage, speed and mileage information of adjacent short strokes.

Optionally, the construction of the classification model of the vehicle running road area using the random forest algorithm includes:

Import the machine learning database; randomly divide the sample data into a training set and a test set, the ratio of the training set to the test set is 7:3, find out the importance of the characteristic parameters, and sort them;

The main parameters of the design classifier are as follows: the number of decision trees in the random forest is set to 178, the maximum tree depth is set to 9, the maximum number of leaf nodes is set to 41, the maximum number of features is set to 31, and the minimum number of samples of leaf nodes is set to 3. The minimum number of samples required for splitting is set to 5;

Adjust the parameters of the model to make the accuracy rate as high as possible, and finally build a classification model of vehicle operating road areas based on machine learning algorithms.

The beneficial effects of the present invention are:

According to geographical location information and road information, the driving conditions of different cities and regions are constructed, which can make the constructed driving conditions more detailed and more representative. The operating area type label of the short-travel data of the present invention is an important condition for determining the driving conditions of the vehicle, and is also the main basis for the synthesis of later characteristic operating conditions. Based on the short-trip database containing the label of the area type to which the vehicle operates, when synthesizing the driving conditions, you can choose to synthesize the driving conditions of typical urban sub-regions according to the regional category, or you can choose to perform it according to the mileage length ratio of the four regional categories. The merging and splicing of the four areas constructs the overall driving conditions of a typical city. Therefore, it is necessary to obtain the operating area type label of the user's short-trip data before the synthesis of driving conditions. When the vehicle is running, the vehicle operating road area classification model of the present invention is used to identify the vehicle operating conditions in real time, which facilitates the real-time adjustment of vehicle control parameters for the vehicle, and timely and accurate calibration of technical parameters to evaluate the performance of the vehicle. Therefore, the vehicle operation road area type identification method of the present invention can not only be used for preliminary preparation of typical working condition development, but also can be used for real-time working condition identification, improve vehicle operation status, and provide data information for vehicle operation cloud service big data analysis.

Description of drawings

Fig. 1 is a flow chart of the vehicle operation road area type identification method of the present invention;

Fig. 2 is a schematic diagram of GPS boundary of road area classification in a typical city in an embodiment of the present invention.

Detailed ways

The present invention will be further described in detail below in conjunction with specific embodiments, which are explanations of the present invention rather than limitations.

Previously, the automotive testing conditions used in the industry were basically international standard working conditions, and the social, economic and geographical characteristics of each city are not the same, and even have great differences. Therefore, typical standard working conditions are not enough to represent different conditions. Actual road driving conditions in the region or city. Not only that, but the road traffic conditions in different geographical areas of the same city are very different. Simple accumulation is not conducive to obtaining typical working conditions with strong adaptability, and cannot more accurately reflect the driving characteristics of specific research objects. Therefore, according to the geographical location information and road information, constructing the driving conditions of different cities and sub-regions can make the constructed driving conditions more detailed and more representative.

First of all, the operating region type label of the short-travel data of the present invention is an important condition for determining the driving conditions of the vehicle, and is also the main basis for the later synthesis of characteristic operating conditions. Based on the short-trip database containing the label of the area type to which the vehicle operates, when synthesizing the driving conditions, you can choose to synthesize the driving conditions of typical urban sub-regions according to the regional category, or you can choose to perform it according to the mileage length ratio of the four regional categories. The merging and splicing of the four areas constructs the overall driving conditions of a typical city. Therefore, it is necessary to obtain the operating area type label of the user's short-trip data before the synthesis of driving conditions.

Secondly, when the vehicle is running, the vehicle operating road area classification model of the present invention is used to identify the vehicle operating conditions in real time, which facilitates real-time adjustment of vehicle control parameters for the vehicle, and timely and accurate calibration of technical parameters to evaluate vehicle performance.

Therefore, the vehicle operation road area type identification method of the present invention can not only be used for preliminary preparation of typical working condition development, but also can be used for real-time working condition identification, improve vehicle operation status, and provide data information for vehicle operation cloud service big data analysis.

The present invention uses the autonomous driving method to obtain original data when constructing vehicle driving conditions, but the road traffic conditions in different geographical areas of typical cities are very different, and simple accumulation is not conducive to obtaining typical working conditions with strong adaptability. However, it cannot more accurately reflect the driving characteristics of a specific research object. Therefore, according to the geographical composition characteristics of typical cities, use arcgis software to divide typical cities into several regions, including urban areas, mountainous areas, suburbs, and expressways; Geographical information and road information are divided, and the "urban" and "suburbs" named by some existing working condition development ideas are the types of working conditions, which are obtained by clustering the short trips of vehicles autonomously driving in the entire typical city , for example, in some existing road condition development ideas, a road condition located in the suburbs may also be classified as "urban" category, the so-called "urban" category actually refers to the vehicle speed is not high, Low-speed working conditions with frequent starts and stops). The driving conditions of vehicles in several areas divided by the present invention all have significant differences. For example, the low-speed operating conditions in the suburban area category are significantly higher than the low-speed operating conditions in the urban area category. Therefore, in the development of typical urban driving conditions, after the typical urban geographical information is divided into several categories, and then the driving conditions are developed, it is beneficial to make the classification of the driving conditions more detailed, typical, and more representative.

The present invention divides the short-distance data into different geographical area types, and based on the short-distance characteristic parameter database containing the area type label to which the vehicle operates, when synthesizing the driving conditions, (1) can choose to synthesize the typical city points according to the area type; Regional driving conditions, (2) You can also choose to merge and splice the four regions according to the mileage length ratio of the four types of regions to construct the total driving conditions of a typical city. This is what step 1, step 2 does.

Step 3 uses the random forest algorithm of machine learning to build a classification model for vehicle running road areas. The training set database for machine learning is: the kinematic feature parameters of the user's short trip + region type labels. The machine learning model will conduct learning and training based on the correlation between the kinematic characteristic parameters of the user's short stroke and the type of area. After training, the user's kinematic characteristic parameters can be input to determine the type of area to which it belongs. In this way, the operating conditions of the vehicle can be identified in real time, which is convenient for real-time adjustment of the vehicle control parameters, and timely and accurate calibration of technical parameters to evaluate the performance of the vehicle.

Innovative selection of machine learning database feature parameters: absolute maximum yaw rate, speed and mileage information of adjacent short strokes, absolute maximum lateral acceleration, braking times, and energy consumption per 100 kilometers.

The specific implementation of the method for identifying the type of vehicle running road area proposed by the present invention mainly includes the following steps:

Step 1. Creation of typical road classification GPS boundary database. Import the national map information data into the GIS platform Arcgis, including provincial administrative regions, county administrative regions, national roads, DEM elevation data, etc. Since every city has a ring expressway or a ring line to divide the suburban area, a typical city’s ring expressway or ring line is used as the boundary between the city and the suburbs, and then intersects with the national road bank to obtain the urban road and the suburbs Road library; expressway library is a collection of high-speed roads in provinces and regions, including suburban expressways and ring expressways; mountainous roads use DEM data to obtain the area of mountainous areas in typical cities in ARCGIS software, thereby obtaining mountainous roads Library; G, S, X, Y, C, Z are used to divide the administrative grades of ordinary roads to distinguish national roads, provincial roads, county roads, township roads, village roads, and special roads, mainly covering suburban roads except expressways and other roads in mountainous areas.

Specifically, in step 1.1, investigate whether the urban-suburban dividing line of each typical city is the ring expressway or the city ring line. Taking Chengdu as an example, Chengdu uses the Fourth Ring Road as the boundary line between the city and the suburbs, and divides the national provincial administrative regions , county-level administrative regions, and national road data are imported into Arcgis, and other provinces and regions other than the provinces and regions where typical cities are deleted are deleted to obtain the vector data of the provinces and regions where typical cities are located. Intersect the provincial data of typical cities with the national county-level administrative district data to obtain the provincial administrative districts of typical cities composed of county-level administrative districts. The redundant counties are deleted according to the existing standard administrative divisions, and then the typical urban administrative divisions composed of county-level administrative divisions are obtained.

Convert the typical urban administrative districts to plane-to-line operations to obtain the outer contour lines of typical urban administrative districts, dots on the contour lines at intervals of 100 meters, and obtain the GPS point information of the outer contour lines of typical urban administrative districts through calculation.

Step 1.2, on the basis of step 1.1, intersect the typical urban administrative districts with the national road data to obtain all road information in a typical city. Copy all the road information in a typical city to obtain the dividing line between the city and the suburbs, open the attribute table of the road information, the name attribute in the attribute table indicates the name of the road, filter out the Chengdu Fourth Ring Road, and then Other unselected roads are deleted, and the suburban boundary is obtained. Observe whether there is a gap in the loop, and if there is a gap, fill it up as a complete coil. Dots are made at intervals of 100 meters on the urban-suburban boundary line, and the GPS point information of the urban-suburban boundary line of typical cities is obtained through calculation. On the basis of obtaining the boundary line, turn the line element into a surface to obtain a typical urban area, copy and cut the typical urban area to obtain suburban and mountainous areas.

Step 1.3: Obtain the information of typical urban highways. On the basis of steps 1.1 and 1.2, intersect the typical urban suburbs and mountainous areas with the national road data to obtain typical urban suburban roads and rename them as expressways to facilitate subsequent operations. Open the attribute table of the road information, filter "fclass" = 'motorway' OR "fclass" = 'motorway_link' by attribute, and get expressways (including ring expressways, ring expressways, ring roads, etc.), expressway interchanges and ramps Wait for the road, switch and delete the rest of the roads to get the expressway database, dot the expressway database at an interval of 100 meters, and obtain the GPS point information of the expressway through calculation. At the same time, the expressways in the urban and suburban road databases are deleted.

Step 1.4, obtain the range of mountainous areas. DEM data is to realize the digital simulation of ground terrain through limited terrain elevation data. The range of mountainous areas will be divided according to the national DEM data. First, according to the latitude and longitude information of typical cities, obtain the surrounding areas of typical cities Multiple blocks of DEM data, import the same layer in Arcgis,

Obtaining a whole piece of merged DEM data through raster mosaic can unify the range of mountain elevation data and facilitate unified processing. The raster data structure is the simplest and most intuitive spatial data structure, also known as the network structure or pixel structure, which refers to dividing the area into grid arrays of equal size and tight arrangement. Defines and includes a code representing the grid's property type or message value.

Then calculate the terrain relief. "1:2.5 million European International Geomorphic Map" and "International Uniform Legend Guide for Medium-Scale Geomorphic Maps" compiled by the International Geographical Union Geomorphological Survey and Cartography Committee, as well as China's "1:4 million Chinese Geomorphic Map", "1:1 million The Standards for Mapping Geomorphological Maps and the Development of Complete Maps of China's Geomorphological Maps all use the degree of topographical relief as the main basis for discriminating the basic morphology of geomorphology. Topographic relief, also known as topographic relief, local topography, and relative height, refers to the height difference between the highest point and the lowest point in a certain area, and its expression is shown in formula (1-1):

R = H _max - H _min (1-1);

In the formula, R is the terrain relief in the analysis area, and H _max and H _min are the maximum and minimum elevation values in the analysis area, respectively.

The entire block of DEM data is subjected to rectangular neighborhood focus statistics, and neighborhood analysis is used to calculate the area around each pixel of statistical data. Define a square neighborhood with a height and width of 23*23 pixels, perform focus statistics on the rectangular block, and obtain the maximum MAXIMUM and minimum MINIMUM, MAXIMUM is the largest value that appears in the calculation neighborhood. MINIMUM is to calculate the minimum value of the cells in the neighborhood. Use the map algebra and raster calculator in the Spatial Analyst tool to calculate "maximum"-"minimum", and the output is a terrain relief map. Use the reclassification tool to reclassify the waviness. Use the identification function to check the approximate stretching value of the boundary line of the mountainous area, divide the upper and lower boundaries into two levels, and obtain the range map of the mountainous area. Since the "natural breaks" categories are based on natural groupings inherent in the data. Class breaks are identified to best group similar values and maximize differences between classes.

First, convert the result of the reclassified raster into a grid to obtain vector data, and then intersect the vector data with the vector data of the suburban area, delete the low-level data, and crop out the mountain range of a typical city.

Handle the tile part. According to the obtained mountain area and the map of China, although the eastern mountain area has ups and downs due to the river, it should belong to the mountain area as a whole, so the tool was used to merge it into a whole area.

Open the editor and select the patch area in the mountain vector data, select the cartography tool→cartographic synthesis→aggregate surface, the input is a mountain area, and when outputting, a new file geodatabase gdb needs to be created in the folder and saved as a mountain aggregation surface; the rest is in In editing, choose to create the automatically generated surface of the element, connect the hole in the large block and the surrounding small area into a whole area and merge it; cut and copy the aggregated surface and the automatically generated surface. The obtained area is the vector data of the mountain area, and the area of the mountain area will be converted to the line to obtain the area line of the mountain area, and the GPS points of the mountain area will be obtained by dots at an interval of 100 meters. In the attribute table information, the tiles will be named according to the block, and the selected part will be excavated Parts are identified with 0 and 1.

Select the intersection and inversion tool, invert the mountainous area in the suburban area, and obtain the ordinary road area, perform the surface-to-line operation on the ordinary road area, and make points on the contour line of the ordinary road area at intervals of 100 meters, and obtain the typical road area through calculation. The GPS point information of the outer contour line of the urban ordinary road area is also named in the information table, and 0 and 1 are used to indicate whether it is the selection range or the removal range.

Since then, the GPS information databases in urban areas, mountainous areas, suburbs, and high-speed areas have all been built, as shown in Figure 2.

Step 2. The user's short-distance travel data is classified by area type. A large amount of user data is used as the input raw data, which includes GPS positioning information, speed, acceleration, SOC, high-voltage power-on signal, motor torque speed and other vehicle operating information. After preprocessing the raw data, according to the conventional early steps of constructing the driving conditions, the user data is divided into several effective segments for short-stroke operation based on the high-voltage power-on signal. Obtain the user's short-stroke database, and calculate the characteristic parameters of each short-stroke for subsequent use.

Construct a function to identify the type of area to which a short trip belongs. This function takes the GPS positioning information of the short trip and the GPS information of various areas of the city (including urban areas, highways, mountainous areas, and suburbs) as input, and outputs the category label of the area to which the short trip belongs. The basic process is as follows:

Since the user's short trip is a running segment, it may span multiple area types, so it is necessary to judge the proportion of the area type to which it belongs based on the longitude and latitude information of the short trip, and define that the longitude and latitude points of the short trip exceed a certain proportion and belong to a certain area type, the short trip is considered to belong to this area type.

To filter out the short trips that belong to the typical urban administrative area, first judge the short trips that belong to the urban area, because the short trip is located in the urban area, it will definitely not belong to the high-speed area. This can reduce the amount of calculation, and the short trip GPS point information, The GPS point information of the urban-suburban boundary is used as input. The first step is to judge whether the GPS point information of the urban-suburban boundary is closed. GPS points in the urban area, and calculate the proportion of GPS points belonging to the urban area in the short trip to the total GPS points. If more than 80% of the GPS points belong to the urban area, then the short trip segment is considered to belong to the urban area. District area category.

There is a difference in judging the short trips belonging to the high-speed area. The judgment method is to find the point closest to each short-trip GPS point in the high-speed library GPS point set, and then calculate the distance between the high-speed library GPS point and the nearest short-trip GPS point according to the latitude and longitude. If the distance is less than 50 meters, the GPS point is considered to belong to the high-speed area. At the same time, the proportion of the GPS points belonging to the high-speed area in the short trip to the total GPS points is calculated. If more than 60% of the GPS points belong to High-speed area, then it is considered that the short-stroke segment belongs to the high-speed area category.

The difference between suburbs and urban areas is that the suburbs are not a convex space, so it is necessary to consider the non-suburbs within the large outline of the suburbs, divide the suburbs and non-suburbs, and use the short-distance GPS point information and the GPS point information of the suburban outline as input , the first step is to judge whether the GPS point information of the suburban contour line is closed, and if it is not closed, make it closed; the second step is to use the inpolygon() function that comes with Matlab to judge the GPS points belonging to the suburban area, and calculate the short The ratio of the GPS points belonging to the suburban area in the itinerary to the total GPS points, if more than 80% of the GPS points belong to the suburban area, then the short trip segment is considered to belong to the suburban area category.

Taking the short-distance GPS point information and the GPS point information of the mountain contour line as input, the first step is to judge whether the GPS point information of the mountain contour line is closed, and if it is not closed, make it closed. The second step is to use the inpolygon that comes with Matlab () function, judge the GPS points belonging to the mountain area, calculate the proportion of the GPS points belonging to the mountain area in the short trip to the total GPS points, if more than 80% of the GPS points belong to the mountain area, then it is considered that the short trip The trip segment belongs to the Mountain Region category.

Then the category of the area that the user passes through in the short trip is obtained, and the label of the type of the area to which the vehicle operates is added to the database of each short trip characteristic parameter for use in subsequent synthetic conditions, and the machine learning database is obtained for use in step 3.

Step 3. Use the random forest algorithm of machine learning to construct a classification model for vehicle running road areas. According to the advantages of the random forest classification application, the random forest algorithm is used to classify and identify the kinematic characteristic parameter data of the short-distance vehicle running based on the characteristic parameter index.

Use the random forest algorithm to train the database to obtain a random forest model. Among them, the random forest model includes multiple decision trees, and the random forest model is a classification recognition model for the user to run machine learning of road area types, and the multiple decision trees are not related to each other. Decision tree is a tree-like structural form, which can be non-binary tree structure or binary tree structure. The random forest model in the embodiment of the present invention is based on the CART decision tree of non-binary tree structure (decision tree algorithm with the reduction of GINI coefficient as the division standard) ) as a weak learner, each decision tree is generated according to the classification regression algorithm. The GINI coefficient, which can represent the frequency at which randomly selected data points in the data set may be misclassified, is calculated as follows:

1. p _k represents the probability that the selected sample belongs to category k, then the probability of this sample being misclassified is 1-p _k ;

2. There are K categories in the sample set, and a randomly selected sample can belong to any one of these k categories, so the sum of the categories For example, there are K categories in the sample set D, and the number of the Kth category is C Gini index of _k :

On the basis of using a decision tree, the random forest randomly selects a part of the sample features on the node, assuming Nsub, and then selects an optimal feature among these randomly selected Nsub sample features to make the left and right subtrees of the decision tree divided. This further enhances the generalization ability of the model. The following is the specific process of using the random forest algorithm to build a vehicle operating road area classification model:

First import the machine learning database, the machine learning database in step 2 contains multiple user short-stroke data, and its characteristic parameters include running time, running distance, running speed, acceleration time, deceleration time, uniform speed time, idle time, acceleration ratio, deceleration Ratio, constant speed ratio, idle speed ratio, maximum speed, average speed, speed standard deviation, maximum acceleration, average acceleration in acceleration section, standard deviation in positive acceleration, maximum deceleration, average deceleration in deceleration section, standard deviation in negative deceleration, average acceleration, Acceleration standard deviation, relative positive acceleration, ratio of different speed intervals, number of accelerations, number of decelerations, number of constant speeds, number of stops, standard deviation of torque, average positive torque, average negative torque, maximum positive torque, maximum negative torque , average torque in idling section, average torque in running section, average positive torque in running section, average negative torque in running section, positive torque time ratio, negative torque time ratio, maximum fluctuation when torque increases, torque decrement Hourly maximum fluctuation, standard deviation of torque fluctuation, average fluctuation when torque increases, average fluctuation when torque decreases, time ratio of high torque interval, time ratio of medium torque interval, time ratio of low torque interval, yaw Absolute maximum value of angular velocity, absolute maximum value of lateral acceleration, number of braking, duration of acceleration segment, energy consumption per 100 kilometers, total damage of motor, total damage of front motor, total damage of rear motor, damage per unit mileage of motor, damage per unit mileage of front motor, The unit mileage damage of the rear motor, the speed and mileage information of adjacent short strokes, and the label of the area type to which the short stroke belongs.

Among them, the selection of characteristic parameters such as the absolute maximum value of the yaw rate, the speed and mileage information of adjacent short strokes, the absolute maximum value of the lateral acceleration, the number of braking times, and the energy consumption per 100 kilometers is conducive to the recognition accuracy of the classification model constructed.

Randomly divide the sample data into a training set and a test set. The ratio of the training set to the test set is 7:3. The importance of the characteristic parameters is calculated and sorted. The main parameters for designing a classifier are as follows:

The number of decision trees in the random forest is set to 178, the maximum tree depth is set to 9, the maximum number of leaf nodes is set to 41, the maximum number of features is set to 31, the minimum number of samples of leaf nodes is set to 3, and the minimum number of samples required for splitting is set to 5.

Here's how to build a random forest:

1. A sample with a sample size of N is drawn N times with replacement, and one sample is drawn each time, and finally N samples are formed. The selected N samples are used to train a decision tree as samples at the root node of the decision tree.

2. When each sample has M attributes, when each node of the decision tree needs to be split, randomly select m attributes from the M attributes, satisfying the condition m≤M. Then adopt some strategy (for example, information gain) from these m attributes to select 1 attribute as the splitting attribute of the node.

3. In the process of forming the decision tree, each node must be split according to step 2 (if the attribute selected by the node next time is the attribute used when the parent node splits, the node has reached the leaf node, and there is no need to continue splitting ), until it can no longer be split. No pruning is performed during the entire decision tree formation process.

4. Build a large number of decision trees according to steps 1 to 3, thus forming a random forest.

Reasonably adjust the ratio of the data in the machine learning database as the training set and the test set, and adjust the model parameters to make the accuracy rate as high as possible. Finally, a vehicle operating road area classification model based on machine learning algorithms is constructed.

It has been verified that the recognition accuracy of the constructed vehicle operating road area classification model is relatively high, and its recognition effect is shown in Table 1.

Table 1

urban area high speed suburbs the mountains test set accuracy 92.85% 94.01% 93.90% 92.21% test set recall 93.97% 94.53% 94.96% 90.63%

In the embodiment of the present invention, step 1 and step 2 construct a short-trip characteristic parameter database containing the type label of the area to which the vehicle operates. Based on this, when synthesizing driving conditions, it is possible to choose to synthesize typical city sub-areas according to the area category. Driving conditions, you can also choose to merge and splice the four regions according to the mileage length ratio of the four types of regions to construct the total driving conditions of a typical city. Step 3 builds a vehicle operating road area classification model based on machine learning, and the type of area to which the vehicle operates can be obtained by inputting the kinematic characteristic parameters and operating data information of the vehicle. Based on this, the vehicle control parameters are adjusted, and timely and accurate Calibrate technical parameters to evaluate vehicle performance.

Matters not covered in the present invention are known technologies.

The above-mentioned embodiments are only to illustrate the technical concept and characteristics of the present invention, and the purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly, and not to limit the protection scope of the present invention. All equivalent changes or modifications made according to the spirit of the present invention shall fall within the protection scope of the present invention.

Claims (7)

1. A vehicle operation road area type identification method, characterized in that, comprising: Step 1: According to the geographical location information, divide the mountain area type in the target area; according to the road type, divide the target area into urban area type, suburban area type and high-speed area type; and represent the range of the area type as discrete latitude and longitude points status information; Step 2: The user data is used as input. The user data includes vehicle operation data information and vehicle GPS positioning information. After preprocessing the data, divide the original data into several kinematic segments according to the short-stroke method, and calculate The characteristic parameters of each short trip are obtained from the short trip characteristic parameter database; according to the vehicle GPS positioning information and the area type range information, the area type to which each short trip belongs is judged; Step 3: In each short-travel characteristic parameter database, add the label of the area type of vehicle operation; at the same time, obtain the machine learning database, use the machine learning algorithm for training, and build a vehicle operation road area classification model based on machine learning. This model will use It is used to identify the road area where the vehicle runs. 2. The vehicle operation road area type identification method according to claim 1, wherein said step 1 specifically includes: According to the scope of urban administrative divisions, determine the boundary line between the urban area and the suburban area; divide the boundary between the city and the suburbs into the urban area, and obtain the GPS point information of the urban area boundary; All expressways in the urban administrative area are classified as high-speed areas, and GPS point information of high-speed areas is obtained; According to the DEM elevation data, calculate the terrain relief, carry out neighborhood analysis, reclassify the terrain relief, obtain the range of mountainous areas, and obtain the GPS point information of the boundary of the mountainous area; Areas other than urban areas, high-speed areas, and mountainous areas are classified as suburban areas, and GPS point information on the borders of suburban areas is obtained. 3 . The method for identifying the type of road area on which vehicles operate according to claim 2 , wherein the boundary line between the urban area and the suburban area is a ring expressway or a ring line. 4 . 4. The vehicle operation road area type identification method according to claim 1, 2 or 3, characterized in that, said step 2 specifically includes: The user's short-stroke data is classified according to the type of area; after preprocessing the original data including vehicle operation data information and vehicle GPS positioning information, the short-stroke is divided according to the short-stroke method, and the original data is divided into several kinematic segments, and the user Short-stroke database, calculating the characteristic parameters of each short-stroke for subsequent use; Construct a function to identify the type of area to which the short trip belongs. The function takes the GPS positioning information of the short trip and the GPS information of the four types of areas as input, and outputs the label of the area type to which the short trip belongs. 5. The vehicle operation road area type identification method according to claim 1, 2 or 3, characterized in that the step three specifically includes: Use the random forest algorithm of machine learning to build a classification model of the vehicle running road area, and obtain the machine learning database, including user short-distance travel data, characteristic parameters and labels, use the random forest algorithm to train the database, obtain the random forest model, and adjust the machine learning reasonably The data in the database is used as the ratio of the training set and the test set. According to the realization effect, the model parameters are adjusted, and finally a machine learning-based vehicle operating road area classification model is constructed. 6. The vehicle operation road area type identification method according to claim 5, wherein the characteristic parameters include: Running time, running distance, running speed, acceleration time, deceleration time, constant speed time, idle time, acceleration ratio, deceleration ratio, uniform speed ratio, idle speed ratio, maximum speed, average speed, speed standard deviation, maximum acceleration, average acceleration in acceleration section , standard deviation of positive acceleration, maximum deceleration, average deceleration of deceleration section, standard deviation of negative deceleration, average acceleration, standard deviation of acceleration, relative positive acceleration, ratio of different speed intervals, number of accelerations, number of decelerations, number of constant speeds, number of stops , standard deviation of torque, average positive torque, average negative torque, maximum positive torque, maximum negative torque, average torque of idle speed segment, average torque of running segment, average positive torque of running segment, average negative rotation of running segment Torque, positive torque time ratio, negative torque time ratio, maximum fluctuation when torque increases, maximum fluctuation when torque decreases, standard deviation of torque fluctuation, average fluctuation when torque increases, average fluctuation when torque decreases amount, time ratio of high torque interval, time ratio of medium torque interval, time ratio of low torque interval, absolute maximum value of yaw rate, absolute maximum value of lateral acceleration, braking times, duration of acceleration segment, energy consumption per 100 kilometers, Motor total damage, front motor total damage, rear motor total damage, motor unit mileage damage, front motor unit mileage damage, rear motor unit mileage damage, speed and mileage information of adjacent short strokes. 7. The vehicle operation road area type identification method according to claim 5, wherein said adopting the random forest algorithm to construct the vehicle operation road area classification model comprises: Import the machine learning database; randomly divide the sample data into a training set and a test set, the ratio of the training set to the test set is 7:3, find out the importance of the characteristic parameters, and sort them; The parameters of the design classifier are as follows: the number of decision trees in the random forest is set to 178, the maximum tree depth is set to 9, the maximum number of leaf nodes is set to 41, the maximum number of features is set to 31, the minimum sample number of leaf nodes is set to 3, and the split The minimum number of samples required is set to 5; Adjust the parameters of the model to make the accuracy rate as high as possible, and finally build a classification model of vehicle operating road areas based on machine learning algorithms.

Images