Bus can bus-based road black point prediction model

Abstract

The application discloses a road black point prediction model based on a bus CAN bus, belongs to the technical field of online road safety online analysis, and specifically comprises the following steps: selecting data, extracting key information features, carrying out road network segmentation, extracting key information, constructing a prediction model, training the ability of the prediction model, carrying out acceptance of the prediction model, and putting into use; based on the training result of the third step, a qualified prediction model is selected, and then the prediction model is put into use. It can combine the historical characteristics of various accidents, extract the weight of the story influencing factor in the historical characteristics, and further predict new black point data, improve the coincidence degree of the prediction result and the actual result in the prediction system, improve the dynamic safety of the predicted bus line, and ensure the practicability of the prediction model.

Description

Technical Field

[0001] This invention relates to the field of online analysis technology for road safety, and more specifically, to a road black spot prediction model based on the CAN bus of buses. Background Technology

[0002] Public transportation, represented by buses, meets the travel needs of most urban residents and plays a vital role in improving the utilization rate of transportation resources and alleviating urban traffic pressure. However, with the advancement of urbanization in my country and the rapid increase in car ownership, more and more road traffic problems are frequently occurring. Therefore, improving the operational safety of bus routes is particularly important.

[0003] Currently, numerous methods exist for identifying accident-prone road sections, each with its own applicability. These methods primarily rely on historical and dynamic traffic accident data to identify and predict accident-prone sections, enabling the identification and warning of accident-prone areas and improving road safety. However, these post-accident identification and control methods often result in significant loss of life and property before the accident identification process is complete. If accident-free accident-prone sections along routes could be predicted and identified, avoidance could be implemented during bus route adjustments or early warnings could be issued on existing routes. This is of great significance for reducing accident rates and ensuring the safety of life and property.

[0004] Currently, CAN bus is one of the most widely used bus technologies in automobiles, and its bus communication data is being explored and applied more and more. One common application is using CAN bus to achieve real-time online monitoring of vehicle faults.

[0005] Using CAN data to study driving behavior is also a common application. By acquiring data from sensors inside the car via the CAN bus, it's possible to determine if the driver is engaging in unsafe driving behaviors, thereby improving driving safety.

[0006] As application exploration deepens, more systematic applications have gradually developed, leading to a vehicle management system based on the CAN bus. This system utilizes the CAN bus to collect data from various vehicle components (motor, electronic control, air conditioning, engine, chassis, body, centralized lubrication) to establish a vehicle management system containing various types of databases. By classifying, summarizing, processing, and analyzing the data from these knowledge bases, key information can be provided for management decisions, comprehensively improving the management level of drivers, vehicles, routes, and safety.

[0007] In China, Yutong, an automotive construction company, has developed a system specifically designed to analyze CAN bus data and provide guidance for vehicle use. This system can effectively monitor and maintain vehicles in real time for customers.

[0008] Based on the above, the inventors discovered that:

[0009] When this prediction system is applied to real-world scenarios, although vehicle management systems have also developed black spot identification and prediction functions, these functions are mainly based on the statistical analysis of dynamic violation data and accident data. However, when making predictions, traditional prediction systems struggle to combine the historical characteristics of different accidents and extract the weights of accident influencing factors from those historical characteristics in order to predict new black spots. This can easily increase the possibility that the prediction results of the prediction system do not match the actual results.

[0010] Therefore, in view of this, we will study and improve the existing structure and provide a road black spot prediction model based on the bus CAN bus, in order to achieve a more practical value. Summary of the Invention

[0011] 1. Technical problems to be solved

[0012] To address the problems existing in the prior art, the purpose of this invention is to provide a road black spot prediction model based on the CAN bus of buses. It can combine the historical characteristics of various accidents and extract the weights of the story influencing factors in the historical characteristics to predict new black spot data, thereby improving the consistency between the prediction results and the actual results in the prediction system. At this time, based on the personalized identification of key feature parameters of the CAN bus with strong correlation to accidents, the black spot road segment prediction model is constructed by inversion, which improves the dynamic safety of predicted bus routes and ensures the practicality of the prediction model.

[0013] 2. Technical Solution

[0014] To solve the above problems, the present invention adopts the following technical solution.

[0015] The road black spot prediction model based on the bus CAN bus is implemented using the following specific steps:

[0016] Step 1: Select data and extract key information features: Based on the bus CAN bus, collect, select and filter CAN bus data. Considering personalized factors, perform personalized cluster analysis on the data and extract key information features.

[0017] Step 2: Perform road network segmentation and extract key information: Based on the road grid information, design a suitable road network segmentation plan, and then combine it with historical black spot data to extract key CAN bus parameter information related to accidents. The key information includes the weight of accident influencing factors.

[0018] Step 3: Construct a prediction model and train its capabilities: Based on the key information features in Step 1 and the historical black spot data in Step 2, combined with the person profile, construct a prediction model. Upon completion of the construction, train the prediction model multiple times to optimize its predictive capabilities.

[0019] Step 4: Accept and put into use the prediction model: Based on the training results of Step 3, select the qualified prediction model, and then put the prediction model into use to conduct a trial run of road safety prediction. After the trial run is qualified, the prediction model will be officially put into use and the acceptance will be completed.

[0020] Furthermore, in step one, when collecting CAN bus data, redundant attribute parameters or low-weight parameter data of the CAN bus data are selected and filtered out to reduce the amount of calculation and lower the difficulty of analysis and calculation.

[0021] Then, two synchronously running analysis lines were established, and cluster analysis of the CAN bus data was performed:

[0022] Analysis line one involves using fuzzy clustering algorithm to cluster vehicle motion states and then performing human profile analysis under vehicle motion states.

[0023] Analysis line two involves using fuzzy clustering algorithm to cluster driving operation states and performing character profile analysis under driving operation states;

[0024] After cluster analysis, fusion features are extracted to complete the extraction of key information features.

[0025] Furthermore, when using the fuzzy clustering algorithm for state clustering, the FCN algorithm is combined for clustering calculation.

[0026] Furthermore, when extracting fusion features, a feature fusion algorithm based on Bayesian theory is used for calculation.

[0027] Furthermore, in step two, when planning the road grid segmentation, a spatial clustering segmentation method is used to divide the road grid.

[0028] Then, historical black spot data is identified, and combined with the key information in step one, driving and operation characteristics are analyzed based on deep information, and key parameter characteristics in the accident-related CAN bus are determined.

[0029] Furthermore, when using spatial clustering to divide the road grid, a sparse subspace clustering algorithm is used for calculation, and the ADMM algorithm is used to solve for the representation matrix.

[0030] Furthermore, when combining historical black spot information and accident-related key CAN bus parameter information, the extracted information includes bus route information, bus operation information, and driving information.

[0031] Furthermore, in step three, the zero-inflation model is first used to analyze the zero-value problem, and the key information from step two is used to construct an initial prediction model.

[0032] Then, dynamic data from the CAN bus and features of the person's profile are collected again to optimize the initial model.

[0033] Furthermore, when training the prediction model, closed roads are used for training, and the number of training iterations is no less than ten.

[0034] Furthermore, in step four, when conducting the acceptance test of the prediction model, a prediction model with a training pass rate of not less than 90% is selected for trial operation of road safety prediction, and then a prediction model with a trial operation pass rate of not less than 98% is selected for acceptance.

[0035] 3. Beneficial effects

[0036] Compared with the prior art, the advantages of this invention are:

[0037] ① This solution, based on the prediction model, is applied in real-world scenarios. This model involves reviewing literature, summarizing and analyzing driving behavior, vehicle status, and traffic accident data analysis methods, constructing a bus operation status representation framework, using qualitative analysis to process the data, separating truth from falsehood, recognizing the inherent laws of things, and using mathematical tools to quantitatively process the research object to more scientifically reveal the laws. Then, using correlation analysis and multiple correspondence analysis, multiple factors are unified, key factors related to danger are extracted, and finally, these key factors are used as input. Multiple algorithms are combined to construct a black spot prediction model. This prediction model can combine the historical characteristics of various accidents and extract the weights of story-related influencing factors from historical characteristics to predict new black spot data, improving the consistency between prediction results and actual results in the prediction system. Furthermore, based on personalized identification of key characteristic parameters of the CAN bus strongly correlated with accidents, a black spot road segment prediction model is constructed, improving the dynamic safety of predicted bus routes and ensuring the practicality of the prediction model.

[0038] ② This scheme, the prediction model of this scheme, studies the characteristic analysis and clustering of bus driving and operation status, which is the basis for carrying out subsequent research. It reduces the amount of data calculation during prediction, reduces the difficulty of analysis and calculation, and selects and optimizes algorithms to adapt to personalized processing needs while taking into account the differences in drivers' physical fitness, personality habits, driving skills and other personal characteristics. Finally, personalized clustering analysis is performed on the filtered data to improve the degree of fit between the prediction model and the drivers.

[0039] ③ In this scheme, the prediction model can select an appropriate algorithm when considering environmental changes, and further analyze the vehicle dynamics and driving behavior characteristics of black spot road sections. It attempts to extract key CAN bus parameters, influence weights and thresholds corresponding to the environment and working conditions, accident types and levels of black spot road sections. This can provide a parameter basis for building a road condition and environmental change monitoring system and a black spot road section prediction model, and improve the accuracy of subsequent predictions.

[0040] ④ This scheme improves the prediction accuracy of traffic accident models when predicting all accidents, including those with minor injuries and no fatalities. Furthermore, it attempts to organically combine fuzzy algorithms, neural networks, and other methods with the prediction model to train and optimize the model, thereby improving the prediction level of risk factors. Attached Figure Description

[0041] Figure 1 This is a schematic diagram illustrating the construction process of the prediction model of the present invention. Detailed Implementation

[0042] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0043] Example 1:

[0044] Please see Figure 1 A road black spot prediction model based on the bus CAN bus is constructed, and the specific implementation steps are as follows:

[0045] Step 1: Select data and extract key information features: Based on the bus CAN bus, collect, select and filter CAN bus data. Considering personalized factors, perform personalized cluster analysis on the data and extract key information features.

[0046] Step 2: Perform road network segmentation and extract key information: Based on the road grid information, design a suitable road network segmentation plan, and then combine it with historical black spot data to extract key CAN bus parameter information related to accidents.

[0047] Step 3: Construct a prediction model and train its capabilities: Based on the key information features in Step 1 and the historical black spot data in Step 2, combined with the person profile, construct a prediction model. Upon completion of the construction, train the prediction model multiple times to optimize its predictive capabilities.

[0048] Step 4: Accept and put into use the prediction model: Based on the training results of Step 3, select the qualified prediction model, and then put the prediction model into use to conduct a trial run of road safety prediction. After the trial run is qualified, the prediction model will be officially put into use and the acceptance will be completed.

[0049] This project follows a scientific research method. First, it obtains data by consulting literature, summarizes and analyzes the data analysis methods of driving behavior, vehicle status and traffic accident data, and constructs a framework for representing the operating status of buses.

[0050] At the same time, qualitative analysis is used to process the materials, distinguish truth from falsehood, and understand the inherent laws of things. Mathematical tools are used to process the research objects quantitatively to reveal the laws more scientifically. Then, correlation analysis and multiple correspondence analysis are used to unify multiple factors, extract key factors related to danger, and finally use key factors as input to construct a black spot prediction model by combining multiple algorithms.

[0051] This prediction model is based on personalized identification of key characteristic parameters of the CAN bus that are strongly correlated with accidents, and inversely constructs a prediction model for black spot road sections, which effectively predicts the dynamic safety of bus routes.

[0052] Example 2:

[0053] Further discussion will be made based on the above embodiment 1.

[0054] See Figure 1 In step one, when collecting CAN bus data, redundant attribute parameters or low-weight parameters of the CAN bus data are selected and filtered out to reduce the amount of calculation and lower the difficulty of analysis and calculation.

[0055] Then, two synchronously running analysis lines were established, and cluster analysis of the CAN bus data was performed:

[0056] Analysis line one involves using fuzzy clustering algorithm to cluster vehicle motion states and then performing human profile analysis under vehicle motion states.

[0057] Analysis line two involves using fuzzy clustering algorithm to cluster driving operation states and performing character profile analysis under driving operation states;

[0058] After cluster analysis, fusion features are extracted to complete the extraction of key information features.

[0059] Based on the above, feature analysis and clustering of bus operation and driving status can be completed to obtain key information and improve the accuracy of the analysis.

[0060] See Figure 1 When using fuzzy clustering algorithm for state clustering, the FCN algorithm is combined for clustering calculation. The specific calculation procedure is as follows:

[0061] %%

[0062] % Implementation of fuzzy C-means clustering method

[0063] %# Import dataset

[0064] clc

[0065] clear

[0066] irisInput = readtable('l\\tsclient\(d\θ-work\learning\fuzzy methods\Iris dataset\iris.txt');

[0067] iris = irisInput(:,2:6);

[0068] % Take a portion of the data

[0069] % Processing datasets;

[0070] x = table2array(iris(1:150,1)); % Converts a table into an array.

[0071] classf = table2ce1l(iris(1:150,5)); % Convert the table into a character array.

[0072] % Encode the type

[0073] y = zeros(150, 1);

[0074] y(strcmp(classf,

[0075] y(strcmp(classf,'versicolor'))=2;

[0076] y(strcmp(classf,'virginica'))=3;

[0077] 2. Set the number of clusters and weighting index. Maximum number of clusters and convergence accuracy.

[0078] n = size(x, 1); % Number of samples

[0079] p = size(x, 2); % Number of variables in a sample

[0080] C = 3;

[0081] % of the items

[0082] m = 1.5;

[0083] The % weighted index m ranges from (1.5, 3.5).

[0084] maxtimes = 158;

[0085] precision = 0.0001;

[0086] %%

[0087] 3. Initialize the membership matrix by taking random numbers in the range [e, 1]. The dimension of the membership matrix is ​​c x n.

[0088] U = rand(c,n);

[0089] % Initialize a random matrix, in the range [e, 1]

[0090] v = zeros(C,D); % A matrix of c rows and p columns, representing the center of each variable in each class.

[0091] J = zeros(200, 1);

[0092] J(1)=e

[0093] errorJ = 1;

[0094] times = 1;

[0095] 5%

[0096] 4. Computing Center

[0097] while(errorJ>precision&× <maxtimes)

[0098] 2. Calculate cluster centers

[0099] times = times + 1;

[0100] fori = 1:3

[0101] v(1,:) = ((U(1,0).^m * x) / sum((U(1,:)).^m));

[0102] nd

[0103] % 5. Update the membership matrix SU._{ij}$S

[0104] % Calculate the Euclidean distance between the sample points and the center points

[0105] d = zeros(c,n);

[0106] % 3x150 matrix

[0107] for i = 1:C

[0108] for j = 1:n

[0109] d(i,j) =

[0110] sum(abs(v(i,:) - x(j,:)));

[0111] % dij refers to the jth sample and

[0112] nd

[0113] % Update the membership matrix. The membership matrix is the extreme point derived by the Lagrange multiplier method

[0114] for i = 1:C

[0115] for j = 1:n

[0116] U(i,j) = sum(d(i,j) / d(:,j))^(-2 / (m - 1));

[0117] end

[0118] end

[0119] u = (mapminmax(U',0,1));

[0120] % Calculate the objective function value J, which is a numerical value

[0121] temp = (U.^m).*(d.^2);

[0122] J(times) = sum(Jtemp(:));

[0123] % Calculate the objective function difference

[0124] errorJ = abs(J(times) - J(times - 1));

[0125] 6. Iterative algorithm, reaching convergence.

[0126] %*7. Display clustering results*

[0127] plot(U(1;:),'k*"');

[0128] holdon;

[0129] plot(U(2;:),'rd');

[0130] holdon;

[0131] plot(U(3;:),'bp');

[0132] G%

[0133] 8. Test Results

[0134] %

[0135] test_x = x(66;:);

[0136] d._test = zeros(C, 1);

[0137] u_test = zeros(C, 1);

[0138] fori = 1:C

[0139] d.test(i,1)=sum(abs(v(i,:)-test.x(1,:)));

[0140] end

[0141] fori = 1:C

[0142] u_test(i,1)=sum(d test(1,1)1d(:,1))^(-2 / (m-1));

[0143] end

[0144] u_test=(mapminmax(U_test',0,1))';

[0145] u_test

[0146] This program segment reduces the computational load of the fuzzy algorithm and eliminates the need for repeated calculations through multiple iterations, thus improving computational efficiency.

[0147] See Figure 1 When extracting fusion features, a feature fusion algorithm based on Bayesian theory is used for calculation, and its calculation expression is as follows:

[0148] First, X represents the classifier output, and w represents the classification label:

[0149] x→ω j

[0150] ifF(ω j )=maxP(ω j x)

[0151]

[0152] Then, to prevent the divisor from being zero, we take the logarithm for calculation:

[0153] Z→ω j

[0154]

[0155] Then, assuming the prior and posterior are approximately equal, the calculation is performed:

[0156] P(ω j x k ,)=P(ω j (1+δ) ki )

[0157] Ultimately, the largest approximate label value will suffice.

[0158] See Figure 1 In step two, when planning the road grid segmentation, the spatial clustering segmentation method is used to divide the road grid.

[0159] Then, historical black spot data is identified, and combined with the key information in step one, driving and operation characteristics are analyzed based on deep information, and key parameter characteristics in the accident-related CAN bus are determined.

[0160] Based on the above, data analysis can be performed on key parameters of historical black spot road sections to improve the scope of data analysis and reduce data analysis errors.

[0161] See Figure 1 When using spatial clustering to divide the road grid, a sparse subspace clustering algorithm is used for calculation, and the ADMM algorithm is used to solve for the representation matrix. The specific expression procedure is as follows:

[0162] Algorithm 1: Sparse Subspace Clustering (SSC) / / SSC is the result of the sparse subspace clustering algorithm.

[0163] Input:A set of points{y}￥1 lying in a union of n linear

[0164] / / Formula Calculation

[0165] Solve the sparse optimization program(5)in the case of uncorrupted data or(13)in the case of corrupted data.

[0166] Normalize the columns of C as / / Formula Calculation

[0167] Form a similarity graph with N nodes representing the data points. Set the weights on the edges between the nodes by

[0168] W = |C| + |C| T / / Formula Calculation

[0169] Apply spectral clustering126] to the similarity graph.

[0170] Output:Segmentation of the data:Y1,Y2,...,Yn.

[0171] The program ultimately yields the cluster partitioning results.

[0172] See Figure 1 When combining historical black spot information and accident-related key CAN bus parameter information, the extracted information includes bus route information, bus operation information, and driving information.

[0173] By combining bus route information, bus operation information, and driving information, information omissions can be avoided during forecasting, thus improving the accuracy of forecasts.

[0174] See Figure 1 In step three, the zero-inflation model is first used to analyze the zero-value problem, and the key information from step two is combined to construct the initial prediction model.

[0175] Then, dynamic data from the CAN bus and features of the person's profile are collected again to optimize the initial model.

[0176] Based on the above, a road black spot prediction model can be constructed, and the optimized prediction model can better meet the prediction requirements.

[0177] See Figure 1 When training the prediction model, closed roads are used for training, and the number of training iterations is no less than ten.

[0178] Closed-road training facilitates management during training, improves safety, and controls the minimum number of training sessions to avoid the randomness of training results.

[0179] See Figure 1 In step four, when conducting the acceptance of the prediction model, a prediction model with a training pass rate of not less than 90% is selected for trial operation of road safety prediction, and then a prediction model with a trial operation pass rate of not less than 98% is selected for acceptance.

[0180] Controlling the pass rate of the prediction model during training and trial operation, while allowing for a certain amount of experimental error, maximizes the effectiveness of the prediction model, thereby ensuring the safety of bus route operation based on the prediction model.

[0181] The above description is merely a preferred embodiment of the present invention; however, the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and its improved concepts, should be covered within the scope of protection of the present invention.

Claims

1. A road black spot prediction model based on bus CAN bus, characterized by: The specific implementation steps for constructing this prediction model are as follows: Step 1: Select data and extract key information features: Based on the bus CAN bus, collect, select and filter CAN bus data. Considering personalized factors, perform personalized cluster analysis on the data and extract key information features. First, when collecting CAN bus data, redundant attribute parameters or low-weight parameters of the CAN bus data are selected and filtered out to reduce the amount of calculation and lower the difficulty of analysis and calculation. Then, two synchronously running analysis lines were established, and cluster analysis of CAN bus data was performed: Analysis line one uses fuzzy clustering algorithm to cluster vehicle motion state and perform human profile analysis under vehicle motion state. Analysis line two involves using fuzzy clustering algorithm to cluster driving operation states and performing character profile analysis under driving operation states; After cluster analysis, fusion features are extracted to complete the extraction of key information features; Step 2: Perform road network segmentation and extract key information: Based on the road grid information, design a suitable road network segmentation plan, and then combine it with historical black spot data to extract key CAN bus parameter information related to accidents. In the process of planning road grid segmentation, spatial clustering segmentation method is used to divide the road grid. Then, historical black spot data is identified, and combined with the key information in step one, driving and operation characteristics are analyzed based on deep information, and key parameter characteristics in the accident-related CAN bus are determined. When using spatial clustering to divide the road grid, a sparse subspace clustering algorithm is used for calculation, and the ADMM algorithm is used to solve it to obtain the representation matrix; Step 3: Construct a prediction model and train its capabilities: Based on the key information features in Step 1 and the historical black spot data in Step 2, combined with the person profile, construct a prediction model. Upon completion of the construction, train the prediction model multiple times to optimize its predictive capabilities. First, a zero-inflation model is used to analyze the zero-value problem, and then the key information from step two is used to construct an initial prediction model. Then, collect dynamic data from the CAN bus and features of the person's profile again to optimize the initial model; When training the prediction model, closed roads are used for training, and the number of training iterations is no less than ten. Step 4: Accept and put into use the prediction model: Based on the training results of Step 3, select the qualified prediction model, and then put the prediction model into use to conduct a trial run of road safety prediction. After the trial run is qualified, the prediction model will be officially put into use and the acceptance will be completed.

2. The road black spot prediction model based on bus CAN bus according to claim 1, characterized in that: When using fuzzy clustering algorithm for state clustering, the FCN algorithm is combined for clustering calculation.

3. The road black spot prediction model based on bus CAN bus according to claim 1, characterized in that: When extracting fusion features, a feature fusion algorithm based on Bayesian theory is used for calculation.

4. The road black spot prediction model based on bus CAN bus according to claim 1, characterized in that: When combining historical black spot information and accident-related key CAN bus parameter information, the extracted information includes bus route information, bus operation information, and driving information.

5. The road black spot prediction model based on bus CAN bus according to claim 1, characterized in that: In step four, when conducting the acceptance of the prediction model, a prediction model with a training pass rate of not less than 90% is selected for trial operation of road safety prediction, and then a prediction model with a trial operation pass rate of not less than 98% is selected for acceptance.

Images