Rail transit driver driving safety risk grade evaluation method and system

Abstract

The invention relates to a rail transit driver driving safety risk grade evaluation method and system. According to the method, after driver portrait data is obtained, the driver portrait data is input into a trained driving safety risk grade evaluation model, and a driving safety risk grade evaluation result is obtained; the training process of the driving safety risk grade evaluation model specifically comprises the steps that an improved DBSCAN clustering method is used for clustering a preprocessed driver portrait set, a clustering result is obtained, and the driver portrait set is obtained in advance; and training a preset classification model based on the clustering result and the original driver portrait to obtain a trained driving safety risk grade evaluation model. Compared with the prior art, the method has the advantages of quickly and accurately realizing automatic driver driving safety risk grade evaluation and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of driver profiling and data analysis, and in particular to a method and system for assessing the driving safety risk level of rail transit drivers. Background Art

[0002] Urban rail transit has outstanding advantages such as large capacity, high efficiency, energy conservation and environmental protection. It is the core of the public transportation system of large cities and an important part of the urban comprehensive transportation system. It plays a key role in the high-quality development of cities.

[0003] Urban rail transit trains are generally driven by a single driver. If the driver's attention is reduced or he falls asleep during the driving process, it is easy to cause a safety accident. Therefore, the driver's driving safety risk assessment is one of the key issues in the organization of rail transit transportation, and the effect of the assessment has a significant impact on safe transportation. The assessment of the driving safety risk level of rail transit drivers is relatively complex and requires a comprehensive assessment from multiple dimensions. It is difficult to match all situations with a set of rules. At present, the main measures to improve train safety include using a highly automated train operation control system to control train operation and developing intelligent monitoring equipment to monitor the train operation status, but there are almost no supervisory and protective measures for the driving risks of train drivers. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for assessing the driving safety risk level of rail transit drivers in order to overcome the defects of the above-mentioned prior art. In view of the problem that there is currently a lack of big data methods for assessing the driving safety risk level of drivers in the field of urban rail transit, this method and system uses the characteristics of the driver's personal portrait to perform dimensionality reduction analysis, cluster analysis, classification analysis and other technologies, divides the driver's driving ability into different levels through clustering, and then trains a classification model based on the clustering results and the original driver portrait, and matches the driver's driving safety level according to the model, etc., to achieve the purpose of assessing the driving safety risk level based on the train driver's portrait.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] According to a first aspect of the present invention, a method for assessing the driving safety risk level of rail transit drivers is provided, comprising the following steps: acquiring driver portrait data; inputting the driver portrait data into a trained driving safety risk level assessment model to obtain a driving safety risk level assessment result; wherein the training process of the driving safety risk level assessment model specifically comprises: clustering a preprocessed driver portrait set using an improved DBSCAN clustering method to obtain a clustering result, wherein the driver portrait set is acquired in advance; based on the clustering result and the driver portrait set, training a preset classification model to obtain a trained driving safety risk level assessment model.

[0007] As a preferred technical solution, the preprocessing process of the driver portrait set includes using a kernel autoencoder to reduce the dimensionality of the input data using a nonlinear dimensionality reduction method.

[0008] As a preferred technical solution, the improved DBSCAN clustering method specifically includes: using the DBSCAN algorithm to identify high-density areas in the pre-processed driver portrait set, eliminating abnormal points, and selecting high-density centers as clustering centers; based on the clustering center points as the number of clustering clusters, re-performing multi-objective optimization radiation diffusion to the surrounding areas to obtain the final clustering results.

[0009] As a preferred technical solution, a preset classification model is trained, and the specific process includes: using a classification algorithm to train the preset classification model and verifying it through a cross-validation method. The classification algorithm includes a support vector machine, a K-nearest neighbor and a neural network.

[0010] As a preferred technical solution, the driver portrait set includes personal portrait data of multiple drivers, and the personal portrait data includes personal basic information, train number, route, number of violations, driving time, fatigue status, driving skill level and historical safety evaluation records.

[0011] According to a second aspect of the present invention, a rail transit driver driving safety risk level assessment system is provided, the system comprising a data acquisition module, a driving safety risk level assessment module, a clustering module and a classification model training module; the data acquisition module is used to acquire driver portrait data; the driving safety risk level assessment module is used to input the driver portrait data into a trained driving safety risk level assessment model to obtain a driving safety risk level assessment result; wherein the training process of the driving safety risk level assessment model is implemented together with the clustering module and the classification model training module, specifically comprising: the clustering module is used to cluster a pre-processed driver portrait set using an improved DBSCAN clustering method to obtain a clustering result, wherein the driver portrait set is pre-acquired; the classification model training module is used to train a preset classification model based on the clustering result and the driver portrait set to obtain a trained driving safety risk level assessment model.

[0012] As a preferred technical solution, the clustering module includes a kernel autoencoder, and the kernel autoencoder is used to perform dimensionality reduction processing on the input data by adopting a nonlinear dimensionality reduction method.

[0013] As a preferred technical solution, the clustering module includes a cluster center determination unit and a diffusion classification unit; the cluster center determination unit is used to use the DBSCAN algorithm to identify high-density areas in the preprocessed driver portrait set, eliminate abnormal points, and select high-density centers as cluster center points; the diffusion classification unit is used to use the cluster center point as the number of cluster clusters, re-perform multi-objective optimization radiation diffusion to the surrounding areas, and obtain the final clustering result.

[0014] As a preferred technical solution, the specific process of the classification model training module training the preset classification model includes: training the preset classification model using a classification algorithm and verifying it through a cross-validation method. The classification algorithm includes a support vector machine, K-nearest neighbor and a neural network.

[0015] As a preferred technical solution, the system further includes a parameter automatic optimization module, and the parameter automatic optimization module is used to adjust the parameters of the clustering module and the classification model training module.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] 1. The method of the present invention adopts an improved DBSCAN clustering method, retains the ability of the DBSCAN algorithm to find core points, modifies its clustering division mode, and can improve the accuracy of clustering compared to the original DBSCAN algorithm. It will not mark a large amount of data as abnormal points, will not cause data errors, and ensure the globality of the data, thereby providing accurate labels for driver portraits and reducing the difficulty of cleaning the training data set;

[0018] 2. The classification algorithm combination provided by the method of the present invention includes three different types of driver portrait-specific classification algorithms, which are suitable for different situations and can meet the risk assessment needs in different scenarios;

[0019] 3. The system of the present invention uses a kernel autoencoder (KAE) for data preprocessing to reduce dimensionality, reducing the data to two dimensions so that the clustering algorithm results can visualize the clustering effect;

[0020] 4. The system of the present invention is equipped with an automatic parameter optimization module, which can quickly obtain calculation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A schematic flow chart of the method provided by the present invention;

[0022] Figure 2 Provide a schematic diagram of the structure of the system for the present invention;

[0023] Among them: 201, data acquisition module; 202, driving safety risk level assessment module; 203, clustering module; 204, classification model training module; 205, management and maintenance module; 206, parameter automatic optimization module;

[0024] Figure 3 The present invention is a flowchart of preprocessing the driver's portrait features in an embodiment of the present invention;

[0025] Figure 4 The following is a schematic diagram of the implementation process of the clustering method provided in the embodiment of the present invention;

[0026] Figure 5 The figure illustrates the training and testing process of the driving safety risk level assessment model provided in the embodiment of the present invention. DETAILED DESCRIPTION

[0027] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0028] Example

[0029] With the continuous development of big data analysis, the assessment of the driving safety risk level of rail transit drivers based on driver portraits has become a problem that needs to be solved, especially in improving the overall safety of the rail transit system. The task of assessing the driving safety risk of rail transit drivers is mainly to explore the association and allocation process between the deep-level information of driver portraits, and its goal is to achieve accurate prediction and risk assessment of driver behavior patterns.

[0030] like Figure 1 and Figure 2 As shown, this embodiment provides a method and system for assessing the driving safety risk level of rail transit drivers. The execution process of the method when applied includes:

[0031] Step S1, obtaining driver portrait data;

[0032] Step S2, inputting the driver portrait data into the trained driving safety risk level assessment model to obtain the driving safety risk level assessment result.

[0033] Among them, the training process of the driving safety risk level assessment model specifically includes:

[0034] Step S21, clustering the preprocessed driver portrait set using an improved DBSCAN clustering method to obtain a clustering result, where the driver portrait set is obtained in advance;

[0035] Step S22: Based on the clustering results, a preset classification model is trained to obtain a trained driving safety risk level assessment model.

[0036] The system is used to implement the aforementioned method, such as Figure 2 As shown, the system includes a data acquisition module 201, a driving safety risk level assessment module 202, a clustering module 203 and a classification model training module 204. The data acquisition module 201 is used to implement step S1, the driving safety risk level assessment module 202 is used to implement step S2, the clustering module 203 is used to implement step S21, and the classification model training module 204 is used to implement step S22.

[0037] Furthermore, the system also includes a management and maintenance module 205 for monitoring and maintaining the entire assessment system. This module can record system operation logs, provide user authority management, and regularly update and optimize the driving safety risk level assessment model to ensure long-term stable operation of the system.

[0038] Furthermore, the system also includes a parameter automatic optimization module 206, which has a parameter automatic optimization function and automatically adjusts the parameters of the clustering and classification algorithms to obtain the best evaluation result.

[0039] The driver portrait set in step S21 includes personal portrait data of multiple drivers. For example, the personal portrait data may include personal basic information, train number, route, number of violations, driving time, fatigue status, driving skill level and historical safety evaluation records. Personal basic information may include age, gender, health status, etc., all of which may affect driving performance; train number and route information help to understand the driver's performance under different conditions, while the number of violations and historical safety evaluation records directly reflect the driver's past driving behavior; monitoring of driving time and fatigue status can reveal whether the driver is at risk of fatigue driving due to long working hours; driving skill level reflects the driver's professional ability level; all of this information combined provides a good data basis for assessing the driver's safety risk level.

[0040] Driver profile features can be manually or automatically selected from these driver profile data for subsequent dimensionality reduction processing. Optionally, driver profile features can be obtained by directly importing a driver profile feature table in excel or csv format, or directly importing it into the system through SQL association query database table. After importing the driver profile features, the system will analyze the driver profile features. You can manually select which specific features need to be analyzed and data processed, or you can select all features for analysis.

[0041] Figure 3One of the processes for preprocessing driver portrait features is shown. First, the input driver portrait features are subjected to standardization processing such as data cleaning; secondly, for the standardized driver portrait data, determine whether dimensionality reduction processing is required: if dimensionality reduction processing is not required, the data is directly used for clustering; if dimensionality reduction processing is required, a dimensionality reduction algorithm is selected. Optionally, a nuclear autoencoder can be used to perform dimensionality reduction processing on the input data using a nonlinear dimensionality reduction method. Specifically, after the user imports the data file, the system will automatically echo the driver portrait features in the file, and can dynamically select features for clustering. The system supports dimensionality reduction of nuclear autoencoders and automatically adapts clustering algorithm parameters.

[0042] When implementing step S21, the clustering module 203 in the system uses the improved DBSCAN clustering method to perform cluster analysis on the driver portrait data to identify driver groups with different driving behavior patterns. The purpose is to mine the deep information of the driver portrait data through clustering, and divide the driver portrait data into different categories to obtain corresponding labels. The divided data is used as the input of the subsequent classification algorithm to ensure that the classification model can accurately identify high-risk and low-risk train driver groups.

[0043] The improved DBSCAN clustering method provided in this embodiment uses the density perception characteristics of the DBSCAN algorithm to perceive the high-density area of ​​data points as the cluster center point, and then based on the cluster center point, radiates and diffuses to the surrounding area to iterate other data points to obtain the final clustering result. The clustering result is also the driver category and its label that are finally divided. The density-based division method is used here to ensure the accuracy and stability of the clustering result. The DBSCAN algorithm first determines a core object, and then identifies the density-reachable objects around the core object according to the set neighborhood radius ε and the minimum number of neighboring points MinPts to form a cluster. In this way, DBSCAN can identify clusters of arbitrary shapes and has good robustness to noise and outliers. Through the optimized cluster center, the data points are re-radiated and diffused, making the final clustering result more accurate.

[0044] Figure 4 One overall implementation process of step S21 is shown. The specific steps are as follows:

[0045] (1) Perform preprocessing operations such as standardization and dimensionality reduction on the selected driver portrait features to improve data quality;

[0046] (2) Initialize all labels to -1, which is the unclustered state;

[0047] (3) Traverse each data point x according to the conditions of equations (1)-(3) i :If the point has been marked, skip it; otherwise, find the Euclidean distance D(xi ,x j ) is less than ε (called neighbors), and the constraints are as follows:

[0048]

[0049] N(xi)={xj|D(xi,xj)<ε} (2)

[0050] |N(xi)|≥MinPts (3)

[0051] In the formula, x represents the data point, MinPts represents the number of points around the data point x, and ε represents the radius of the circle with the core point as the center.

[0052] If the neighbor point N(x i ) is less than MinPts, the point is marked as -2 (noise point); otherwise, a new cluster is created and the point and its neighbors are added to the cluster. The label assignment conditions are as follows:

[0053]

[0054] Where labels[i] represents the label of each point.

[0055] This step must generate at least one valid cluster (excluding noise). If no cluster is formed or all points are marked as noise points, the ε and MinPts parameter space are readjusted.

[0056] (4) Obtain the initial cluster labels and calculate the cluster center using the density algorithm. The mathematical model is added based on DBSCAN as follows:

[0057]

[0058] In the formula, C j is the set of points of the jth cluster, |C j | is the size of the cluster.

[0059] (5) Identify the center point {μ1, μ2, ..., μ k} is qualified as the initial center, that is, whether the cluster center is in a high-density area. If not, fine-tune ε and MinPts and re-execute the density algorithm until the center point meets the requirements.

[0060] (6) Calculate the distance from each data point to all center points. The mathematical model is as follows:

[0061]

[0062] Each data point is then assigned to the nearest center point and the label is updated, expressed as:

[0063]

[0064] In the formula, D(x i ,μ j ) is each data point x i To each cluster center μ j distance, ω1, ω2 are weights, Var(C j ) is the intra-class variance within the jth cluster.

[0065] (7) Set the algorithm termination conditions. The algorithm convergence conditions are as follows:

[0066]

[0067] Where T is the tolerance.

[0068] Repeat steps (6) to (7) until the center point converges, that is, the change is less than the tolerance set in advance, or the maximum number of iterations is reached. When the termination condition is met, the final driver category can be obtained.

[0069] When the classification model training module 204 in the system implements step S22, the clustering results and the original driver portrait set are used as training data sets. The system supports the use of three machine learning algorithms, namely support vector machine (SVM), K-nearest neighbor (KNN), and neural network (NN), to train the classification model. After the training is completed, the system will verify the model through methods such as cross-validation to ensure that the model has good generalization ability and accuracy. Among them, this embodiment provides three classification algorithms for selection, which can effectively train the classification model to meet the risk assessment needs in different scenarios. Specifically, the first classification algorithm is support vector machine (SVM), which finds the optimal decision boundary in high-dimensional space and is suitable for the classification of nonlinear problems; the second classification algorithm is K-nearest neighbor (KNN), which classifies by calculating the distance between the test sample and the known category sample, and is suitable for the rapid classification of small-scale data sets; the third classification algorithm is a neural network (NN) classification algorithm, which automatically learns the internal structure and pattern of the driver portrait and continuously adjusts the connection weights and thresholds between neurons to make the model more refined.

[0070] Figure 5 The training and testing process of the driving safety risk level assessment model provided in this embodiment is shown, that is, one implementation process of step S22, and the specific steps are as follows:

[0071] (8) Based on the driver portrait data and the driver category data (and its cluster labels) finally divided in step (7), the training set and the test set are divided, the cluster labels generated in step (7) are used as target labels, and different classification algorithms are used to divide the data into training set and test set;

[0072] (9) Select a classification algorithm. Select an appropriate classification algorithm, such as support vector machine (SVM), K-nearest neighbor (KNN), and neural network (NN), to train the cluster label data;

[0073] (10) Classification model training: train different preset classification models through training data to learn the feature representation of each category;

[0074] (11) Model verification and optimization: Use the test set to verify the performance of the classification model and adjust the model parameters according to evaluation indicators such as accuracy, recall rate, and F1 score to improve the classification effect. The optimization process can be achieved through methods such as cross-validation or grid search.

[0075] After the above training and testing process, the final trained driving safety risk level assessment model is obtained.

[0076] Automated driver driving safety risk level assessment: In actual application, that is, when the system implements step S1 and step S2, the specific process includes:

[0077] The data acquisition module 201 first receives the input real-time or historical driver portrait data and sends it to the driving safety risk level assessment module 202. Then the driving safety risk level assessment module 202 inputs the new real-time or historical driver portrait data into the aforementioned trained driving safety risk level assessment model to obtain the driving safety risk level assessment result, thereby realizing the automatic driver driving safety risk level assessment quickly and accurately. The assessment result can be directly linked to the driver's safe driving file, providing decision support for rail transit operating organizations, helping them to take corresponding safety measures, such as arranging additional training, adjusting work shifts, etc., to reduce potential safety risks.

[0078] In summary, the rail transit driver driving safety risk level assessment method and system provided by the present invention can effectively assess the driver's driving safety risk, provide strong technical support for the safe operation of rail transit, and have the following advantages in practical application:

[0079] 1. Supplement the driver-level monitoring and supervision that is most commonly used in the field of urban rail transit, which can realize the assessment of the driving safety risk level of rail transit drivers and supplement the safety of rail transit operations.

[0080] 2. By real-time monitoring of the driver's driving behavior and physiological state, potential safety hazards can be discovered in a timely manner, so that preventive measures can be taken to reduce the risk of accidents.

[0081] 3. Ability to provide personalized safety advice and training programs based on the driver’s personal driving habits and historical performance to improve the driver’s overall driving level and safety awareness.

[0082] 4. Accurate risk assessment and timely intervention will help build a safer and more efficient rail transit operation environment.

[0083] 5. Comprehensively consider vehicle information, location information, driver information, driving information and other features to build a comprehensive risk assessment model. Through this comprehensive assessment, the system can identify high-risk behavior patterns that may lead to accidents and issue alerts to drivers or managers in a timely manner.

[0084] 6. Ability to dynamically adjust risk assessment criteria according to different operating environments and conditions to ensure the accuracy and applicability of assessment results.

[0085] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for evaluating the driving safety risk level of a rail transit driver, characterized in that: The following steps are involved: Obtain driver profile data; Inputting the driver portrait data into a trained driving safety risk level assessment model to obtain a driving safety risk level assessment result; The training process of the driving safety risk level assessment model specifically includes: Using an improved DBSCAN clustering method to cluster the preprocessed driver portrait set to obtain a clustering result, wherein the driver portrait set is obtained in advance; Based on the clustering results and the driver portrait set, a preset classification model is trained to obtain a trained driving safety risk level assessment model.

2. The method for evaluating the driving safety risk level of rail transit drivers according to claim 1 is characterized in that: The preprocessing process of the driver portrait set includes using a kernel autoencoder to perform dimensionality reduction processing on the input data using a nonlinear dimensionality reduction method.

3. The method for evaluating the driving safety risk level of rail transit drivers according to claim 1 is characterized in that: The improved DBSCAN clustering method specifically includes: Using the DBSCAN algorithm to identify high-density areas in the preprocessed driver portrait set, remove abnormal points, and select high-density centers as cluster centers; Based on the cluster center point as the number of clusters, multi-objective optimization radiation diffusion is performed again to the surroundings to obtain the final clustering result.

4. The method for evaluating the driving safety risk level of rail transit drivers according to claim 1, characterized in that: Training the preset classification model. The specific process includes: The preset classification model is trained using a classification algorithm and verified using a cross-validation method. The classification algorithm includes a support vector machine, a K-nearest neighbor, and a neural network.

5. The method for evaluating the driving safety risk level of rail transit drivers according to claim 1 is characterized in that: The driver portrait set includes personal portrait data of multiple drivers, and the personal portrait data includes personal basic information, train number, route, number of violations, driving time, fatigue status, driving skill level and historical safety evaluation records.

6. A rail transit driver driving safety risk level assessment system, characterized in that: The system includes a data acquisition module, a driving safety risk level assessment module, a clustering module and a classification model training module; The data acquisition module is used to acquire driver portrait data; The driving safety risk level assessment module is used to input the driver portrait data into a trained driving safety risk level assessment model to obtain a driving safety risk level assessment result; The training process of the driving safety risk level assessment model is implemented by using a clustering module and a classification model training module, and specifically includes: The clustering module is used to cluster the pre-processed driver portrait set using an improved DBSCAN clustering method to obtain a clustering result, wherein the driver portrait set is obtained in advance; The classification model training module is used to train a preset classification model based on the clustering results and the driver portrait set to obtain a trained driving safety risk level assessment model.

7. The rail transit driver driving safety risk level assessment system according to claim 6 is characterized in that: The clustering module includes a kernel autoencoder, and the kernel autoencoder is used to perform dimensionality reduction processing on input data by adopting a nonlinear dimensionality reduction method.

8. The rail transit driver driving safety risk level assessment system according to claim 6 is characterized in that: The clustering module includes a cluster center determination unit and a diffusion classification unit; The cluster center determination unit is used to identify high-density areas in the pre-processed driver portrait set using the DBSCAN algorithm, remove abnormal points, and select high-density centers as cluster center points; The diffusion classification unit is used to perform multi-objective optimization radiation diffusion to the surroundings based on the cluster center point as the number of cluster clusters to obtain the final clustering result.

9. The rail transit driver driving safety risk level assessment system according to claim 6 is characterized in that: The specific process of the classification model training module training the preset classification model includes: The preset classification model is trained using a classification algorithm and verified using a cross-validation method. The classification algorithm includes a support vector machine, a K-nearest neighbor, and a neural network.

10. The rail transit driver driving safety risk level assessment system according to claim 6, characterized in that: The system also includes a parameter automatic optimization module, which is used to adjust the parameters of the clustering module and the classification model training module.

Images