A vehicle driving condition identification method based on XGBoost

Abstract

The present invention relates to a vehicle driving condition identification method based on XGBoost. First, the effective historical driving data of the vehicle is obtained from the database of the Internet of Vehicles service cloud platform, and processed to form a vehicle driving data sample suitable for the XGBoost classification algorithm model; then the XGBoost classification recognition model is prepared with a training data set, and the initial data set suitable for the unsupervised learning algorithm is formed after processing; the driving condition type label of each sample data in the initial data set is analyzed and determined to form a classification recognition model training data set; the XGBoost classification recognition model is trained with a data set with a driving condition category label, and the accuracy rate reaches the optimal after hyperparameter adjustment; finally, the trained XGBoost vehicle driving condition identification model is used to identify the vehicle driving data sample, and the driving condition type is counted. The present invention is simple and efficient, has a strong generalization ability, can be deployed and applied on a variety of platforms, and is conducive to actual production applications.

Description

Technical Field

[0001] The present invention relates to a vehicle driving condition recognition method, in particular to a vehicle driving condition recognition method based on XGBoost, and belongs to the technical field of intelligent networked vehicle data analysis. Background Art

[0002] At present, the number of cars in China is constantly increasing, and the data of Internet of Vehicles is also growing explosively. These data are like huge mineral resources. Using data analysis, machine learning, and deep learning to extract valuable information from a large amount of Internet of Vehicles data is of great significance for improving users' safe travel behavior and vehicle energy conservation and environmental protection.

[0003] The driving condition reflects the curve of the speed change of the car over time during the driving process. Through this curve, we can see the type of road environment the car is in during the operation. The driving condition of the car affects the safety and economic performance of the car. Real-time analysis of the working condition of the car is of great reference significance to various automobile manufacturers in the process of developing new energy vehicles. It provides rich research materials for the study of strategies such as unmanned driving. At the same time, by studying the driving condition of the car, it is conducive to designing more accurate energy management strategies and enhancing the fuel utilization rate of the car.

[0004] Therefore, in the process of developing the functions of the automobile remote monitoring and management system and designing the control strategy of the vehicle controller, it is an urgent problem to mine and analyze the automobile operation data and introduce the driving condition information into the monitoring and management system and vehicle control strategy to improve vehicle driving safety and energy saving and environmental protection. To this end, it is particularly important to propose a vehicle driving condition identification method based on XGBoost. Summary of the invention

[0005] The purpose of the present invention is to address the problems existing in the prior art and propose a vehicle driving condition identification method based on XGBoost, which can identify and count the vehicle driving conditions, make full use of the performance of various hardware facilities of existing vehicles, and further study and analyze the vehicle driving road environment type, realize the functional development of the automobile remote monitoring management system and design a well-designed vehicle controller control strategy.

[0006] The specific technical solution of the present invention is as follows: A vehicle driving condition identification method based on XGBoost comprises the following specific steps:

[0007] Step 1, obtaining the valid historical driving data of the vehicle for 24 hours on a certain day from the enterprise's Internet of Vehicles service cloud platform database. The driving history data is the data information that can be collected by Iveco's existing vehicle network hardware equipment, including data collection time, vehicle speed, cumulative mileage, longitude, latitude, atmospheric pressure (kPa), engine net output torque (%), engine speed (rpm), engine fuel flow (L / h), reactant residue (%), intake air volume (kg / h), SCR inlet temperature (℃), DPF pressure difference (kPa), engine coolant temperature (℃), fuel tank level (%), etc. The present invention mainly requires the three types of data collection time, vehicle speed and longitude and latitude position information.

[0008] Step 2, processing the historical driving data in the previous step to form a vehicle driving data sample suitable for the XGBoost classification algorithm model;

[0009] Step 3: Prepare a training data set for the XGBoost classification and recognition model by collecting road data from a certain area or using standard vehicle driving cycle data to form an initial data set suitable for an unsupervised learning algorithm.

[0010] Step 4, after the initial data set formed in the above steps is learned by an unsupervised learning algorithm, the driving condition type label of each sample data in the initial data set is analyzed and determined to form a classification recognition model training data set;

[0011] Step 5: Use the data set with driving condition category labels to train the XGBoost classification recognition model. After hyperparameter adjustment, the accuracy rate reaches the optimal value.

[0012] Step 6: Use the trained XGBoost vehicle driving condition recognition model to identify the vehicle driving data samples and count the driving condition types.

[0013] Furthermore, in step 2, the processing steps of historical driving data are specifically as follows:

[0014] Step 2.1: The historical driving data obtained in step 1 is expressed as Data = [T Veh_v Lon Lat], where the vehicle data collection time T = [t 1 t 2 …t n ], collect vehicle status information once every 1s;

[0015] The vehicle running status information is the vehicle speed Veh_v = [v 1 v 2 …v n ];

[0016] The location information is longitude Lon = [lon1 lon 2 …lon n ], latitude Lat = [lat 1 lat 2 …lat n ];

[0017] Step 2.2, the driving data Data to be identified is processed in blocks, and the driving data Data is divided into windows starting from the time point when the first speed is not 0 and the fixed driving time t is used to obtain the driving data condition blocks;

[0018] Step 2.3, obtain acceleration information from the speed information difference of each working condition block;

[0019] Step 2.4, calculate the characteristic parameter information of each working condition block, the characteristic parameter information includes 15 types, including average vehicle speed, average vehicle speed during operation, maximum vehicle speed, speed standard deviation, average acceleration, maximum acceleration, average deceleration, maximum deceleration, acceleration standard deviation, deceleration standard deviation, acceleration time ratio, deceleration time ratio, idling time percentage, cruising time percentage and high-speed time ratio;

[0020] Step 2.5, the calculation formulas for each characteristic parameter information are:

[0021] Average speed:

[0022] Average speed during operation:

[0023] Maximum speed: v max =max(v 1 ,v 2 ,v 3 ,v 4 ,…,v n );

[0024] Speed ​​standard deviation:

[0025] Average acceleration:

[0026] Maximum acceleration: a max =max(a 1 ,a 2 ,a 3 ,a 4 ,…,a n );

[0027] Average deceleration:

[0028] Maximum deceleration: d max =max(d1 ,d 2 ,d 3 ,d 4 ,…,d n );

[0029] Acceleration standard deviation:

[0030] Standard deviation of deceleration:

[0031] Acceleration time ratio:

[0032] Deceleration time ratio:

[0033] Idle time percentage:

[0034] Percentage of cruising time:

[0035] High-speed time ratio:

[0036] Step 2.6, all the working condition blocks calculate the fifteen characteristic parameters in step 2.5, and then express them as: Pre_Data = [v ave v av v max v m a av a max d av d max a m d m a p d p S p V p h p ].

[0037] Furthermore, in step 3, the specific steps of preparing the training data set for the XGBoost classification recognition model are:

[0038] Step 3.1, collect road data of a certain area or use standard vehicle driving cycle data as initial model training data;

[0039] Step 3.2: The model training data is processed by windowing in step 2, and further compound partitioning is performed to increase the number of data set samples and reduce the event contingency and randomness of the training data to obtain the training data Data Training ;

[0040] Step 3.3: For the training data Data obtained in step 3.2Training Perform unsupervised learning, obtain the true labels of the training samples, and normalize the training samples. The calculation formula is:

[0041]

[0042] Step 3.4: Perform data dimensionality reduction on the data after normalization in step 3.3. The data dimensionality reduction methods include at least principal component analysis, autoencoder, linear discriminant analysis, singular value decomposition and local linear embedding method. Select appropriate dimensionality reduction methods to perform data dimensionality reduction on the data to reduce the research complexity of the model and finally obtain the dimensionality reduction data Data. Dimensionality Reduction ;

[0043] Step 3.5, perform cluster analysis on the dimensionality reduction data in the previous step to find the category labels in the data. According to SSE and clustering silhouette coefficient and combined with the daily driving environment type, the driving conditions are divided into four categories, namely urban congestion, urban, suburban, and high-speed driving conditions. The K-Means++ clustering algorithm is used for cluster analysis. In order to correspond to the above four types of driving conditions, the initial clustering categories are set to four. At the beginning of clustering, the number of clusters needs to be defined, and then clustering is performed according to the four types. After clustering, the final cluster center is numerically analyzed to determine which specific driving condition type each category corresponds to, and the true label of each data sample is obtained, and the data label is returned to the training data Data in step 3.2 Training In the end, each sample in the training data corresponds to a driving condition type label.

[0044] Furthermore, in the step 3.1, the roads in a certain area are areas with a certain terrain trend, as well as areas where vehicles are actually driven for a long time during production, and data is obtained through collection; the vehicle standard driving cycle condition data is an internationally accepted vehicle testing standard driving condition.

[0045] Furthermore, in step 4, the specific steps of analyzing the driving condition type labels in the training data set are:

[0046] Step 4.1: The labels after cluster analysis in step 3.5 are returned to the training data Training In the process, the returned data labels are used for classification, and the average value of 15 characteristic parameters in each type of data is calculated as the final cluster center;

[0047] Step 4.2: Analyze and compare the data of the four cluster centers, and assign corresponding types to each data label through analysis and comparison.

[0048] Furthermore, in step 5, the XGBoost classification recognition model training includes:

[0049] Step 5.1, Data Training The data set is divided into 80% as training set and 20% as validation set;

[0050] Step 5.2: Use the training set to train the XGBoost machine learning model. Use the CART regression tree model. You can use Python3 to call the Sklearn library, use the XGBoost module for model training, or use the Xgboost library native interface for calling and learning.

[0051] Step 5.3, determine the learning rate and the number of decision trees, select the learning rate learning_rate. Generally, the learning rate is 0.1. However, for different problems, the ideal learning rate sometimes fluctuates between 0.05 and 0.3. Select the ideal number of decision trees corresponding to this learning rate. Here, select the learning rate of 0.1.

[0052] Step 5.4, after the given learning rate and number of decision trees, perform decision tree specific parameter tuning, the maximum depth of the tree max_depth, usually between 3 and 10; the minimum leaf node sample weight min_child_weight; the minimum loss function drop value gamma required for node splitting, the default is 0; the random sampling ratio subsample, usually between 0.5 and 1; the proportion of columns sampled by each tree colsample_bytree, usually between 0.5 and 1. In the process of determining a tree, give each hyperparameter an initial value based on the empirical value, and use grid search and three-fold cross validation to achieve parameter optimization;

[0053] Step 5.5, tuning of XGBoost regularization parameters, lambda and alpha parameters, reducing the complexity of the model, thereby improving the generalization ability of the model;

[0054] Step 5.6, reduce the learning rate learning_rate = 0.01, adjust the number of decision trees, repeat steps 5.4 and 5.5 to further determine the ideal parameters.

[0055] Furthermore, in step 6, the XGBoost driving condition identification model application step includes:

[0056] Step 6.1, obtain the Pre_Data data in step 2.6, use the machine learning model XGBoost driving condition recognition model to identify it, and return the data label;

[0057] In step 6.2, the returned data label is mapped to the corresponding driving condition type information, and the driving condition type is statistically analyzed to obtain the proportion data of each type. The type proportion can also be used to show that the vehicle is approaching a certain driving condition over a period of time.

[0058] Compared with the prior art, the beneficial effects of the present invention are as follows: a vehicle driving condition identification method based on XGBoost proposed in the present invention analyzes the user's historical driving data, identifies and counts the vehicle driving conditions, provides data support for realizing user travel safety analysis and monitoring, and makes full use of existing vehicle-mounted equipment and Internet of Vehicles cloud platform database and other conditions. Through a set of identification and statistical methods, the driving condition identification and statistics of all vehicle travel data in the enterprise Internet of Vehicles cloud platform can be realized or used as key information to adjust vehicle control strategies. The method is simple and efficient, has strong generalization ability, can be deployed and applied on a variety of platforms, and is conducive to actual production applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The present invention will be further described below in conjunction with the accompanying drawings.

[0060] Figure 1 This is a flow chart of the vehicle driving condition identification and statistical method based on XGBoost described in the present invention.

[0061] Figure 2 It is a flow chart for processing the historical driving data described in the present invention.

[0062] Figure 3 It is a schematic diagram of the composite partitioning principle of the working condition blocks of the training data set described in the present invention.

[0063] Figure 4 It is a flow chart of the driving condition type label analysis of the training data described in the present invention.

[0064] Figure 5 It is the PCA principal component analysis contribution rate and cumulative contribution rate diagram described in the present invention.

[0065] Figure 6 This is a curve diagram of SSE changing with clustering category K described in the present invention.

[0066] Figure 7 It is a diagram of the shadow area of ​​the silhouette coefficient and the average silhouette coefficient when K=4 described in the present invention.

[0067] Figure 8 It is a diagram of the numerical analysis results of the cluster centers described in the present invention.

[0068] Fig. 9 It is a flow chart of the XGBoost classification model training described in the present invention.

[0069] Fig.10 It is a flow chart of the driving condition identification record and statistics described in the present invention. DETAILED DESCRIPTION

[0070] Example

[0071] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with the accompanying drawings and specific implementation methods. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0072] like Figure 1 As shown, the vehicle driving condition identification method based on XGBoost includes the following steps:

[0073] (1) Obtain the vehicle's valid 24-hour driving history data on a specific day from the enterprise's Internet of Vehicles service cloud platform database. The driving history data mainly includes data collection time, vehicle speed, and latitude and longitude location information;

[0074] (2) Processing historical driving data to form vehicle driving data samples suitable for the XGBoost classification algorithm model;

[0075] (3) Preparation of training data set for XGBoost classification and recognition model: by collecting data from roads in a certain area or using standard vehicle driving cycle conditions, the data is processed to form an initial data set suitable for unsupervised learning algorithm;

[0076] (4) After the unsupervised learning algorithm of the initial data set is learned, the driving condition type label of each sample data in the initial data set is analyzed and determined to form a classification recognition model training data set;

[0077] (5) The XGBoost classification and recognition model is trained using a dataset with driving condition category labels. After hyperparameter adjustment, the accuracy is optimized.

[0078] (6) Use the trained XGBoost vehicle driving condition recognition model to identify vehicle driving data samples and count the driving condition types.

[0079] like Figure 2 As shown, the step (2) of processing historical driving data includes:

[0080] (2.1) The data obtained in step (1) is expressed as Data = [T Veh_v Lon Lat], where the vehicle data collection time T = [t 1 t 2 …t n ], collect vehicle status information once every 1s;

[0081] Vehicle running status information: driving speed Veh_v = [v 1 v 2 …v n ];

[0082] Location information: Longitude Lon = [lon 1 lon 2 …lon n ], latitude Lat = [lat 1 lat 2 …lat n ];

[0083] (2.2) The driving data Data to be identified is processed in blocks. The driving data Data is divided into windows starting from the time point when the first speed is not 0 and the fixed driving time t is used to obtain the driving data condition blocks;

[0084] (2.3) Obtain acceleration information from the speed information difference of each working condition block;

[0085] (2.4) Calculate the characteristic parameter information of each working condition block, including average vehicle speed, average vehicle speed during operation, maximum vehicle speed, speed standard deviation, average acceleration, maximum acceleration, average deceleration, maximum deceleration, acceleration standard deviation, deceleration standard deviation, acceleration time ratio, deceleration time ratio, idling time percentage, cruising time percentage, and high-speed time ratio, a total of 15 characteristic parameters;

[0086] (2.5) The characteristic parameter information calculation formula is:

[0087] Average speed:

[0088] Average speed during operation:

[0089] Maximum speed: v max =max(v 1 ,v 2 ,v 3 ,v 4 ,…,v n );

[0090] Speed ​​standard deviation:

[0091] Average acceleration:

[0092] Maximum acceleration: a max =max(a 1 ,a 2 ,a 3 ,a 4 ,…,a n);

[0093] Average deceleration:

[0094] Maximum deceleration: d max =max(d 1 ,d 2 ,d 3 ,d 4 ,…,d n );

[0095] Acceleration standard deviation:

[0096] Standard deviation of deceleration:

[0097] Acceleration time ratio:

[0098] Deceleration time ratio:

[0099] Idle time percentage:

[0100] Percentage of cruising time:

[0101] High-speed time ratio:

[0102] (2.6) All the working condition blocks calculate the 15 characteristic parameters in (2.5) and then express them as:

[0103] Pre_Data = [v ave v av v max v m a av a max d av d max a m d m a p d p S p V p h p ]

[0104] like Figure 4 As shown, the step (3) XGBoost classification model training data preparation step includes:

[0105] (3.1) Collecting road data from a certain area or using vehicle standard driving cycle data as initial model training data;

[0106] (3.2) The roads in a certain area are areas with a certain terrain trend, as well as areas where vehicles travel for a long time during actual production, and data is collected;

[0107] (3.3) The vehicle standard driving cycle data described above is the internationally accepted vehicle test standard driving condition;

[0108] (3.4) The model training data is windowed in step (2) and further compound divided, such as Figure 3 As shown, the number of data set samples is increased, and the event contingency and randomness of the training data are reduced to obtain the training data Data Training ;

[0109] (3.5) Unsupervised learning, obtain the true label of the training sample, normalize the training sample, and calculate the formula as follows:

[0110]

[0111] (3.6) After the normalization in step (3.5), the data is subjected to data dimensionality reduction. Commonly used data reduction methods include principal component analysis (PCA), autoencoder (AE), linear discriminant analysis (LDA), singular value decomposition (SVD), local linear embedding (LLE) and other methods. Select appropriate data reduction methods to reduce the complexity of the model research. Here, the principal component analysis method is used for data dimensionality reduction, such as Figure 5 As shown, it can be generally considered that the first n principal components with a cumulative contribution rate greater than 80% can express the representative information of the initial data set. For this reason, the first four principal components with a cumulative contribution rate greater than 85% are selected here to represent the initial training data, and the dimension reduction data Data is obtained. Dimensionality Reduction .

[0112] (3.7) Perform cluster analysis on the reduced-dimensional data to find the category labels in the data, such as Figure 6 and 7As shown, according to the SSE and clustering silhouette coefficient curves, it can be seen that when the cluster category K = 4, the SSE curve has an inflection point. Using the elbow rule, K = 4 is the optimal number of cluster categories, and the silhouette coefficient area shadows for each category exceed the sample average silhouette coefficient dotted line, so the number of clusters is reasonable. Further combined with the daily driving environment type, the driving conditions can be divided into four categories, namely urban congestion, urban, suburban, and high-speed driving conditions. Currently, the commonly used clustering algorithms are based on partitioning K-Means, K-Medoids, K-Modes, K-Medians, Kernel K-Means, based on hierarchical BIRCH, CURE, CHAMELEON, based on density DBSCAN, OPTICS, DENCLU, etc. Considering the data samples and classification tasks, the K-Means++ clustering algorithm is used for clustering analysis. The initial clustering categories are four, and the true label of each data sample is obtained. The data label is returned to step (3.4) training data Data Training So far, each sample in the training data corresponds to a driving condition type label.

[0113] like Figure 5 As shown, the step (4) of analyzing the driving condition type labels in the training data set includes:

[0114] (4.1) After cluster analysis in step (3.7), the labels are returned to the training data Training In , the returned data labels are used for classification, and the average value of 15 characteristic parameters in each type of data is calculated as the final cluster center, such as Figure 8 As shown in the figure, after the obtained cluster center analyzes the numerical values ​​of each characteristic parameter, it maps the corresponding driving condition type label. For the high-speed condition, it can be analyzed from the figure that its maximum speed of 100.4600km / h and average speed of 83.770km / h are the largest compared with the other three types, and the acceleration-related characteristics are relatively the smallest, such as the maximum acceleration of 0.8155 and the minimum deceleration of -3.5118, that is, the speed fluctuation is smaller than the other three types, the cruising time accounts for 0.98, the largest proportion, and the idling time accounts for 0.0105, the smallest proportion. The congestion condition is the opposite, and the other two conditions are between high speed and congestion.

[0115] (4.2) Analyze and compare the data of the four cluster centers, and assign corresponding types to each data label through analysis and comparison.

[0116] like Fig. 9 As shown, the step (5) XGBoost classification recognition model training includes:

[0117] (5.1) Data TrainingThe data set is divided into 80% as training set and 20% as validation set;

[0118] (5.2) Use the training set to train the XGBoost machine learning model. Xgboost is a boosted tree model that integrates many tree models to form a strong classifier. The tree model used is the CART regression tree model. You can use Python to call the Sklearn library, use the XGBoost module for model training, or use the Xgboost library native interface for calling and learning;

[0119] (5.3) Determine the learning rate and the number of decision trees. Select the learning rate learning_rate. Generally, the learning rate is 0.1. However, for different problems, the ideal learning rate sometimes fluctuates between 0.05 and 0.3. Select the ideal number of decision trees corresponding to this learning rate. Here, the learning rate is 0.1.

[0120] (5.4) After the learning rate and number of decision trees are given, the specific parameters of the decision tree are tuned. The maximum depth of the tree is usually between 3 and 10; the minimum leaf node sample weight is min_child_weight; the minimum loss function drop value gamma required for node splitting is 0 by default; the proportion of random sampling is usually between 0.5 and 1; the proportion of the number of columns randomly sampled per tree is usually between 0.5 and 1. In the process of determining a tree, an initial value is given to each hyperparameter based on the empirical value, and the grid search and three-fold cross validation are used to achieve parameter optimization;

[0121] (5.5) Tuning of XGBoost regularization parameters, lambda and alpha parameters, reducing the complexity of the model, thereby improving the generalization ability of the model;

[0122] (5.6) Reduce the learning rate learning_rate = 0.01, adjust the number of decision trees, repeat steps (5.4) and (5.5) to further determine the ideal parameters.

[0123] like Fig.10 As shown, the application steps of the XGBoost driving condition identification model include:

[0124] (6.1) Obtain the Pre_Data data in step (2), identify it using the machine learning model XGBoost driving condition recognition model, and return the data label;

[0125] (6.2) The returned data labels are mapped to the corresponding driving condition type information. The driving condition types are statistically analyzed to obtain the proportion data of each type. The type proportion can also be used to show whether the vehicle is approaching a certain driving condition over a period of time.

[0126] In specific production deployment applications, the present invention can run in an online cloud platform server user terminal or in a flexible application deployment, and can be applied to a variety of service frameworks. As a simple and efficient analysis and identification method tool, it calls the enterprise's internal Internet of Vehicles service cloud platform database or vehicle terminal to perform real-time analysis and identification, and transmits the identification results to the Internet of Vehicles remote monitoring service cloud platform or vehicle controller, thereby realizing the analysis and tracking of the user's driving conditions of the vehicle, ensuring the driver's driving safety, or using it as key information to formulate a good control strategy and improve vehicle energy conservation and environmental protection.

[0127] In general, the present invention provides a simple, efficient, flexible and generalizable algorithm recognition model, which utilizes a large amount of vehicle driving data for analysis, obtains information on the driving condition of the vehicle when the driver is driving the vehicle from a large amount of user driving data, and explores the changing patterns of the driving environment exhibited by the vehicle during driving, so as to subsequently develop vehicle and user portrait models in combination with data mining, machine learning, deep learning and other technologies, and design a more complete vehicle control strategy based on the information on vehicle driving condition changes; for vehicles that are identified to be in a specific driving condition for a long time, risk management measures such as safety education recommendation programs and UBI auto insurance plans can be formulated for the vehicle drivers, so as to improve road traffic safety and reduce enterprise operating risks.

[0128] In addition to the above examples, the present invention may also have other implementations. Any technical solution formed by equivalent replacement or equivalent transformation falls within the protection scope of this claim.

Claims

1. A vehicle driving condition identification method based on XGBoost, characterized by: The specific steps include: Step 1: Obtain the valid historical driving data of the vehicle for 24 hours on a certain day from the existing Internet of Vehicles service cloud platform database, wherein the historical driving data at least includes data collection time, vehicle speed, and latitude and longitude location information; Step 2, processing the historical driving data in the previous step to form a vehicle driving data sample suitable for the XGBoost classification algorithm model; Step 3: Prepare a training data set for the XGBoost classification and recognition model by collecting road data from a certain area or using standard vehicle driving cycle data, and then process it to form an initial data set suitable for the unsupervised learning algorithm. The specific steps for preparing the training data set for the XGBoost classification and recognition model are as follows: Step 3.1, collect road data of a certain area or use standard vehicle driving cycle data as initial model training data; Step 3.2: The model training data is processed by windowing in step 2, and further compound partitioning is performed to increase the number of data set samples and reduce the event contingency and randomness of the training data to obtain the training data Data Training ; Step 3.3: For the training data Data obtained in step 3.2 Training Perform unsupervised learning, obtain the true labels of the training samples, and normalize the training samples. The calculation formula is: Step 3.4: Perform data dimensionality reduction on the data after normalization in step 3.

3. The data dimensionality reduction methods include at least principal component analysis, autoencoder, linear discriminant analysis, singular value decomposition and local linear embedding method. Select appropriate dimensionality reduction methods to perform data dimensionality reduction on the data to reduce the research complexity of the model and finally obtain the dimensionality reduction data Data. DimensionalityReduction ; Step 3.5, perform cluster analysis on the dimensionality reduction data in the previous step to find the category labels in the data. According to SSE and clustering silhouette coefficient and combined with the daily driving environment type, the driving conditions are divided into four categories, namely urban congestion, urban, suburban, and high-speed driving conditions. The K-Means++ clustering algorithm is used for cluster analysis. In order to correspond to the above four types of driving conditions, the initial clustering categories are set to four. At the beginning of clustering, the number of clusters needs to be defined, and then clustering is performed according to the four types. After clustering, the final cluster center is numerically analyzed to determine which specific driving condition type each category corresponds to, and the true label of each data sample is obtained, and the data label is returned to the training data Data in step 3.2 Training In the end, each sample in the training data corresponds to a driving condition type label; Step 4: After the initial data set formed in the above steps is learned by the unsupervised learning algorithm, the driving condition type label of each sample data in the initial data set is analyzed and determined to form a classification recognition model training data set; the specific steps of analyzing the driving condition type label in the training data set are: Step 4.1: The labels after cluster analysis in step 3.5 are returned to the training data Training In the process, the returned data labels are used for classification, and the average value of 15 characteristic parameters in each type of data is calculated as the final cluster center; Step 4.2, analyze and compare the data of the four cluster centers, and assign corresponding types to each data label through analysis and comparison; Step 5: Use the data set with driving condition category labels to train the XGBoost classification recognition model. After hyperparameter adjustment, the accuracy rate reaches the optimal value. Step 6: Use the trained XGBoost vehicle driving condition recognition model to identify the vehicle driving data samples and count the driving condition types.

2. The vehicle driving condition identification method based on XGBoost according to claim 1, characterized in that: In step 2, the processing steps of historical driving data are specifically as follows: Step 2.1: The historical driving data obtained in step 1 is expressed as Data = [T Veh_v Lon Lat], where the vehicle data collection time T = [t1 t2 … t n ], the vehicle status information is collected once every 1s; the vehicle running status information is the driving speed Veh_v=[v1 v2 … v n ]; The location information is longitude Lon = [lon1 lon2 ... lon n ], latitude Lat = [lat1 lat2 … lat n ]; Step 2.2, the driving data Data to be identified is processed in blocks, and the driving data Data is divided into windows starting from the time point when the first speed is not 0 and the fixed driving time t is used to obtain the driving data condition blocks; Step 2.3, obtain acceleration information from the speed information difference of each working condition block; Step 2.4, calculate the characteristic parameter information of each working condition block, the characteristic parameter information includes 15 types, including average vehicle speed, average vehicle speed during operation, maximum vehicle speed, speed standard deviation, average acceleration, maximum acceleration, average deceleration, maximum deceleration, acceleration standard deviation, deceleration standard deviation, acceleration time ratio, deceleration time ratio, idling time percentage, cruising time percentage and high-speed time ratio; Step 2.5, the calculation formulas for each characteristic parameter information are: Average speed: Average speed during operation: Maximum speed: v max =max(v1,v2,v3,v4,…,v n ); Speed ​​standard deviation: Average acceleration: Maximum acceleration: a max =max(a1,a2,a3,a4,…,a n ); Average deceleration: Maximum deceleration: d max =max(d1,d2,d3,d4,…,d n ); Acceleration standard deviation: Standard deviation of deceleration: Acceleration time ratio: Deceleration time ratio: Idle time percentage: Percentage of cruising time: High-speed time ratio: Step 2.6, all the working condition blocks calculate the fifteen characteristic parameters in step 2.5 and then express them as: Pre_Data=[v ave v av v max v m a av a max d av d max a m d m a p d p S p V p h p ]。 3. The vehicle driving condition identification method based on XGBoost according to claim 1, characterized in that: In the step 3.1, the road in a certain area is an area with a certain terrain trend, and an area where the vehicle is driven for a long time during the actual production process, and data is obtained through collection; The vehicle standard driving cycle data is the internationally accepted vehicle testing standard driving condition.

4. The vehicle driving condition identification method based on XGBoost according to claim 1, characterized in that: In step 5, the XGBoost classification recognition model training includes: Step 5.1, Data Training The data set is divided into 80% as a training set and 20% as a validation set; Step 5.2: Use the training set to train the XGBoost machine learning model. Use the CART regression tree model. You can use Python3 to call the Sklearn library, use the XGBoost module for model training, or use the Xgboost library native interface for calling and learning. Step 5.3, determine the learning rate and the number of decision trees, select the learning rate learning_rate. Generally, the learning rate is 0.

1. However, for different problems, the ideal learning rate sometimes fluctuates between 0.05 and 0.

3. Select the ideal number of decision trees corresponding to this learning rate. Here, select the learning rate of 0.

1. Step 5.4, after the given learning rate and number of decision trees, perform decision tree specific parameter tuning, the maximum depth of the tree max_depth, between 3 and 10; the minimum leaf node sample weight min_child_weight; The minimum loss function drop value gamma required for node splitting is 0 by default; the proportion of random sampling subsample is between 0.5 and 1; the proportion of the number of columns sampled by each tree colsample_bytree is between 0.5 and 1. In the process of determining a tree, an initial value is given to each hyperparameter based on the empirical value, and the grid search and three-fold cross validation are used to achieve parameter optimization; Step 5.5, tuning of XGBoost regularization parameters, lambda and alpha parameters, reducing the complexity of the model, thereby improving the generalization ability of the model; Step 5.6, reduce the learning rate learning_rate = 0.01, adjust the number of decision trees, repeat steps 5.4 and 5.5 to further determine the ideal parameters.

5. The vehicle driving condition identification method based on XGBoost according to claim 1, characterized in that: In step 6, the XGBoost driving condition identification model application step includes: Step 6.1, obtain the Pre_Data data in step 2.6, use the machine learning model XGBoost driving condition recognition model to identify it, and return the data label; In step 6.2, the returned data label is mapped to the corresponding driving condition type information, and the driving condition type is statistically analyzed to obtain the proportion data of each type. The type proportion can also be used to show that the vehicle is approaching a certain driving condition over a period of time.

Images