Intelligent traffic monitoring system accident identification model training method

Abstract

The invention relates to an accident identification model training method of an intelligent traffic monitoring system, multifunctional intelligent sensing equipment in the intelligent traffic monitoring system adopts a three-axis acceleration sensor to monitor expressway vibration information, and the collected data is uploaded to a cloud data server for road accident identification model training. Through data enhancement, the problems of insufficient training sample data and non-representativeness of the sample data when a road accident identification model is trained due to the fact that data sets of accidents are few and collected data sets are concentrated are solved, and accident sample data enhancement is performed by adopting a variational auto-encoder, so that the accuracy of road accident identification is improved. And the purpose of rapidly training an iterative traffic monitoring system accident identification model is achieved.

Description

Technical Field

[0001] The invention belongs to the field of public transportation road monitoring, and in particular relates to an accident identification model training method for an intelligent traffic monitoring system. Background Art

[0002] As the mileage of highways continues to increase, the number of cars continues to increase, and economic development increases people's travel needs, various types of accidents on highways are emerging one after another. It is feasible to add intelligent traffic monitoring modules to the highways to collect vibration information on the road and then judge the state of the road based on the vibration data.

[0003] However, the roads are divided into multi-lane and single-lane roads, and the types of vehicles on the roads include small family cars, large commercial muck trucks, transport trucks, etc. Therefore, the accuracy of judging whether a fault occurs on the road only by the vibration threshold is low. For example, when the vibration amplitude is large, it is impossible to judge whether there is a slight collision between vehicles or a bump in the muck truck. Therefore, a training method for accident recognition model of intelligent traffic monitoring system is proposed. Summary of the invention

[0004] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0005] In view of the following technical problems in the prior art: the roads are divided into multi-lane and single-lane roads, and the types of vehicles on the roads include small family cars, large commercial muck trucks, transport trucks, etc. Therefore, the accuracy of judging whether a fault occurs on the road only by the vibration threshold is low. For example, when the vibration amplitude is large, it is impossible to judge whether there is a slight collision between vehicles or a bump in the muck truck.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for training an accident recognition model of an intelligent traffic monitoring system, the training method comprising the following steps:

[0007] S1: Use multifunctional intelligent sensing equipment to collect data to form a data set;

[0008] S2: Obtain the characteristics of the collected data using variational mode decomposition;

[0009] S3: Using support vector machine to train accident recognition model;

[0010] S4: Optimize the parameters in the support vector machine using the sparrow optimization algorithm;

[0011] S5: Output the accident identification model in the intelligent traffic monitoring system.

[0012] As a preferred technical solution for the accident recognition model training method of an intelligent traffic monitoring system, due to the small number of accident data sets, the collected data is concentrated, which leads to insufficient training sample data and unrepresentative sample data when training the road accident recognition model. A variational autoencoder is now used to enhance the accident sample data.

[0013] As a preferred technical solution for the accident identification model training method of an intelligent traffic monitoring system, the intelligent traffic monitoring system places multifunctional intelligent sensing equipment on different types of roads such as expressways, provincial roads, and national roads, as well as single-lane, double-lane, three-lane, and four-lane roads, to collect a data set lasting one year. The data set is divided into a training data set and a test data set.

[0014] As an optimal technical solution for the accident identification model training method of an intelligent traffic monitoring system, the multifunctional intelligent sensing device in the intelligent traffic monitoring system uses a three-axis acceleration sensor to monitor the vibration information of the highway, and the collected data is uploaded to the cloud data server for road accident identification model training;

[0015] Road accidents include: road collapse, road construction, vehicle-to-vehicle collision, vehicle-to-guardrail collision, icy roads and normal conditions.

[0016] As a preferred technical solution for the accident identification model training method of an intelligent traffic monitoring system, the vibration information collected by the multifunctional intelligent sensing equipment is a time-series three-axis acceleration signal. The signal is selected by a sliding window method, and the window size is twice the signal period.

[0017] As a preferred technical solution for an accident identification model training method of an intelligent traffic monitoring system, a variational mode decomposition algorithm is used to extract features from data containing vibration information, such as x-axis amplitude, x-axis frequency band, y-axis amplitude, y-axis frequency band, z-axis amplitude and z-axis frequency band.

[0018] As an optimal technical solution for the accident recognition model training method of an intelligent traffic monitoring system, a variational autoencoder is used to perform data enhancement on training sets and test sets of different types of accidents.

[0019] As a preferred technical solution for the accident recognition model training method of an intelligent traffic monitoring system, when training a road accident recognition model, a support vector machine algorithm is used to map sample data including amplitude and frequency characteristics to a high-dimensional feature space and a training model.

[0020] As an optimal technical solution for the accident identification model training method of an intelligent traffic monitoring system, two key parameters need to be set when training the support vector machine model, namely, the kernel parameter and the penalty factor;

[0021] The kernel parameters and penalty factors are optimized through the swarm intelligence optimization algorithm Sparrow algorithm.

[0022] As an optimal technical solution for the accident identification model training method of an intelligent traffic monitoring system, the sparrow algorithm optimization steps are as follows:

[0023] S1: Initialize the sparrow population, as shown in formula (1):

[0024]

[0025] In the formula, d represents the number of parameters to be optimized. In the present invention, the parameters to be optimized are the kernel parameter and the penalty factor. Therefore, d=2, and n is the number of sparrows in the population. In the present invention, the number of population initialization is set to 100.

[0026] S2: Calculate the fitness of each individual in the sparrow population. In the present invention, the fitness is the accuracy of the test results of the test set;

[0027] S3: According to the fitness of the sparrows in the group, the sparrows with higher fitness are the discoverers, and the rest are the followers;

[0028] The finder is responsible for searching for food and providing directions for other sparrows to search for food;

[0029] Followers mainly follow discoverers to search for food; sentinels are responsible for discovering predators and sending danger information to the sparrow population, so that the sparrow population abandons food and migrates to a safe area; if the number of discoverers in the sparrow population is set to account for 20% of the total population, the sparrows will be sorted according to their individual fitness, and the sparrows with the top 20% fitness will be discoverers;

[0030] S4: Update the position of the discoverer in the sparrow population, which can be expressed as:

[0031]

[0032] Where t represents the number of iterations, i and j represent the subscript and superscript in formula (1), i represents the i-th sparrow, j represents the j-th dimension parameter of the i-th sparrow, itermax represents the maximum number of iterations, which can be expressed as T; Q is a random number that obeys the normal distribution; L is a matrix whose elements are all 1; R2 represents the safety value of the sparrow population position, and ST is the warning value of the sparrow population position;

[0033] α is a uniform random number between 0 and 1, R2 is a uniform random number between 0 and 1, and ST is a constant between 0.5 and 1, which can be set during the initialization phase;

[0034] When the safety value R2 is less than the warning value ST, it means that the position of the sparrow population is safe, and the search position of the discoverer in the population is larger; when the safety value R2 is less than the warning value ST, it means that the position of the sparrow population is unsafe, and the sparrow population should migrate;

[0035] S5: Update the position of the follower in the sparrow population, which can be expressed as:

[0036]

[0037] In the formula, is the optimal position of the discoverer in the sparrow population at the current iteration, x worst is the worst position of an individual in the global sparrow population, A is a 1×d matrix whose elements are randomly selected from -1 or 1, A + It is expressed as:

[0038] A + =A T (AA T ) -1 (4)

[0039] S6: Update the position of the sentinel in the sparrow population, which can be expressed as:

[0040]

[0041] In the formula, represents the optimal position in the sparrow population at the tth iteration, β is the step size, and the value is a random number. The random number is normally distributed with a mean of 0 and a variance of 1;

[0042] f i is the fitness of the current iteration number of the i-th sparrow, f g 、f w is the optimal fitness and the worst fitness before the t+1th iteration;

[0043] ε is a very small constant to avoid f i =f w , which makes formula (5) meaningless;

[0044] Formula (5) shows that when the fitness of the vigilant is equal to f g When , it indicates that the sparrow that has been warned is in the best position, and the sparrow will flee to the nearby area;

[0045] If the fitness of the sentinel is not equal to f g, then the alerter escapes to the current optimal position;

[0046] The swarm intelligence algorithm sparrow algorithm is used to optimize the kernel parameters and penalty factors in the support vector machine.

[0047] The beneficial effects of the accident recognition model training method for an intelligent traffic monitoring system of the present invention are as follows: through data enhancement: solving the problem that the data set of accidents is small, resulting in the concentration of collected data, and then leading to insufficient training sample data and unrepresentative sample data when training the road accident recognition model, and using variational autoencoders to enhance the accident sample data, thereby achieving the purpose of fast training and iterative traffic monitoring system accident recognition model;

[0048] By optimizing the parameters in the support vector machine through the sparrow optimization algorithm, vibration data enhancement of different accident types is achieved, so that the accident recognition model of the traffic monitoring system can more accurately identify and distinguish accident types. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0050] Figure 1 It is the algorithm flow chart of the present invention;

[0051] Figure 2 It is a classification diagram of the data set of the present invention;

[0052] Figure 3 The sliding window of the present invention;

[0053] Figure 4 is a flow chart of the variational mode decomposition algorithm of the present invention;

[0054] Figure 5 It is the principle diagram of the variational autoencoder of the present invention;

[0055] Figure 6 This is the sparrow optimization algorithm diagram of the present invention.

[0056] N is 20, n is the number of iterations of optimization using the sparrow optimization algorithm;

[0057] The input signal x is the vibration information collected during different types of accidents, and the output signal y is the generated vibration information of the corresponding accident type, thereby realizing vibration data enhancement of different accident types;

[0058] The window length is L, and the step size of each window movement is M. DETAILED DESCRIPTION

[0059] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0060] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0061] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0062] Secondly, the present invention is described in detail with reference to the schematic diagram. When describing the embodiments of the present invention in detail, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0063] Example:

[0064] like Figure 1 As shown, the present invention proposes a method for training an accident recognition model of an intelligent traffic monitoring system, and the training method comprises the following steps:

[0065] S1: Use multifunctional intelligent sensing equipment to collect data to form a data set;

[0066] S2: Obtain the characteristics of the collected data using variational mode decomposition;

[0067] S3: Using support vector machine to train accident recognition model;

[0068] S4: Optimize the parameters in the support vector machine using the sparrow optimization algorithm;

[0069] S5: Output the accident identification model in the intelligent traffic monitoring system.

[0070] like Figure 2 As shown in the figure, data enhancement: Due to the small number of accident data sets, the collected data is concentrated, which leads to insufficient training sample data and unrepresentative sample data when training the road accident recognition model. Variational autoencoder is now used to enhance the accident sample data.

[0071] The intelligent traffic monitoring system places multifunctional intelligent sensing devices on different types of roads, such as expressways, provincial roads, and national roads, and on single-lane, double-lane, three-lane, and four-lane roads. It collects a data set for one year, which is divided into a training data set and a test data set.

[0072] The multifunctional intelligent sensing device in the intelligent traffic monitoring system uses a three-axis acceleration sensor to monitor the vibration information of the highway. The collected data is uploaded to the cloud data server for road accident recognition model training;

[0073] Road accidents include: road collapse, road construction, vehicle-to-vehicle collision, vehicle-to-guardrail collision, icy roads and normal conditions.

[0074] like Figure 3 As shown, the vibration information collected by the multifunctional intelligent sensing device is a time-series three-axis acceleration signal. The signal is selected by a sliding window method, and the window size is twice the signal period.

[0075] like Figure 4 As shown, the data containing vibration information uses a variational mode decomposition algorithm to extract features in the data, such as x-axis amplitude, x-axis frequency band, y-axis amplitude, y-axis frequency band, z-axis amplitude and z-axis frequency band.

[0076] like Figure 5 As shown in the figure, variational autoencoders are used to perform data augmentation on training sets and test sets of different types of accidents.

[0077] like Figure 6 As shown, when training the road accident recognition model, the support vector machine algorithm is used to map the sample data including amplitude and frequency features to the high-dimensional feature space and train the model.

[0078] When training the support vector machine model, two key parameters need to be set, namely the kernel parameter and the penalty factor;

[0079] Optimize kernel parameters and penalty factors through swarm intelligence optimization algorithm Sparrow algorithm;

[0080] Sparrow algorithm optimization steps:

[0081] S1: Initialize the sparrow population, as shown in formula (1):

[0082]

[0083] In the formula, d represents the number of parameters to be optimized. In the present invention, the parameters to be optimized are the kernel parameter and the penalty factor. Therefore, d=2, and n is the number of sparrows in the population. In the present invention, the number of population initialization is set to 100.

[0084] S2: Calculate the fitness of each individual in the sparrow population. In the present invention, the fitness is the accuracy of the test results of the test set;

[0085] S3: According to the fitness of the sparrows in the group, the sparrows with higher fitness are the discoverers, and the rest are the followers;

[0086] The finder is responsible for searching for food and providing directions for other sparrows to search for food;

[0087] Followers mainly follow discoverers to search for food; sentinels are responsible for discovering predators and sending danger information to the sparrow population, so that the sparrow population abandons food and migrates to a safe area; if the number of discoverers in the sparrow population is set to account for 20% of the total population, the sparrows will be sorted according to their individual fitness, and the sparrows with the top 20% fitness will be discoverers;

[0088] S4: Update the position of the discoverer in the sparrow population, which can be expressed as:

[0089]

[0090] Where t represents the number of iterations, i and j represent the subscript and superscript in formula (1), i represents the i-th sparrow, j represents the j-th dimension parameter of the i-th sparrow, itermax represents the maximum number of iterations, which can be expressed as T; Q is a random number that obeys the normal distribution; L is a matrix whose elements are all 1; R2 represents the safety value of the sparrow population position, and ST is the warning value of the sparrow population position;

[0091] α is a uniform random number between 0 and 1, R2 is a uniform random number between 0 and 1, and ST is a constant between 0.5 and 1, which can be set during the initialization phase;

[0092] When the safety value R2 is less than the warning value ST, it means that the position of the sparrow population is safe, and the search position of the discoverer in the population is larger; when the safety value R2 is less than the warning value ST, it means that the position of the sparrow population is unsafe, and the sparrow population should migrate;

[0093] S5: Update the position of the follower in the sparrow population, which can be expressed as:

[0094]

[0095] In the formula, is the optimal position of the discoverer in the sparrow population at the current iteration, x worst is the worst position of an individual in the global sparrow population, A is a 1×d matrix whose elements are randomly selected from -1 or 1, A + It is expressed as:

[0096] A + =AT (AA T ) -1 (4)

[0097] S6: Update the position of the sentinel in the sparrow population, which can be expressed as:

[0098]

[0099] In the formula, represents the optimal position in the sparrow population at the tth iteration, β is the step size, and the value is a random number. The random number is normally distributed with a mean of 0 and a variance of 1;

[0100] f i is the fitness of the current iteration number of the i-th sparrow, f g 、f w is the optimal fitness and the worst fitness before the t+1th iteration;

[0101] ε is a very small constant to avoid f i =f w , which makes formula (5) meaningless;

[0102] Formula (5) shows that when the fitness of the vigilant is equal to f g When , it indicates that the sparrow that has been warned is in the best position, and the sparrow will flee to the nearby area;

[0103] If the fitness of the sentinel is not equal to f g , then the alerter escapes to the current optimal position;

[0104] Using swarm intelligence algorithm sparrow algorithm to optimize kernel parameters and penalty factors in support vector machine

[0105] It will be appreciated that in the development of any actual implementation, as in any engineering or design project, numerous implementation-specific decisions may be made. Such a development effort may be complex and time-consuming, but will be a routine task of design, fabrication, and production for those of ordinary skill having the benefit of this disclosure without undue experimentation.

[0106] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for training an accident identification model for an intelligent traffic monitoring system, characterized in that: The training method includes the following steps: S1: Use multifunctional intelligent sensing equipment to collect data to form a data set; S2: Obtain the characteristics of the collected data using variational mode decomposition; S3: Using support vector machine to train accident recognition model; S4: Optimize the parameters in the support vector machine using the sparrow optimization algorithm; S5: Output the accident identification model in the intelligent traffic monitoring system.

2. The method for training an accident identification model for an intelligent traffic monitoring system according to claim 1, characterized in that: Due to the small number of accident data sets, the collected data is concentrated, which leads to insufficient training sample data and unrepresentative sample data when training the road accident recognition model. Variational autoencoder is now used to enhance the accident sample data.

3. The method for training an accident identification model for an intelligent traffic monitoring system according to claim 2, characterized in that: The intelligent traffic monitoring system places multifunctional intelligent sensing equipment on different types of roads, such as expressways, provincial roads, and national roads, as well as single-lane, double-lane, three-lane, and four-lane roads, to collect a one-year data set, which is divided into a training data set and a test data set.

4. The method for training an accident identification model for an intelligent traffic monitoring system according to claim 3, characterized in that: The multifunctional intelligent sensing device in the intelligent traffic monitoring system uses a three-axis acceleration sensor to monitor the vibration information of the highway. The collected data is uploaded to the cloud data server for road accident recognition model training; Road accidents include: road collapse, road construction, vehicle-to-vehicle collision, vehicle-to-guardrail collision, icy roads and normal conditions.

5. The method for training an accident identification model for an intelligent traffic monitoring system according to claim 3, characterized in that: The vibration information collected by the multifunctional intelligent sensing device is a time-series three-axis acceleration signal. The signal is selected by a sliding window method, and the window size is twice the signal period.

6. The method for training an accident identification model for an intelligent traffic monitoring system according to claim 5, characterized in that: The data containing vibration information uses a variational mode decomposition algorithm to extract features in the data, such as x-axis amplitude, x-axis frequency band, y-axis amplitude, y-axis frequency band, z-axis amplitude and z-axis frequency band.

7. The method for training an accident identification model for an intelligent traffic monitoring system according to claim 1, characterized in that: Variational autoencoders are used to perform data augmentation on training sets and test sets of different types of accidents.

8. The method for training an accident identification model for an intelligent traffic monitoring system according to claim 1, characterized in that: When training the road accident recognition model, the support vector machine algorithm is used to map the sample data including amplitude and frequency features into a high-dimensional feature space and train the model.

9. The method for training an accident identification model for an intelligent traffic monitoring system according to claim 1, characterized in that: When training the support vector machine model, two key parameters need to be set, namely the kernel parameter and the penalty factor; The kernel parameters and penalty factors are optimized through the swarm intelligence optimization algorithm Sparrow algorithm.

10. The method for training an accident identification model for an intelligent traffic monitoring system according to claim 9, characterized in that: Sparrow algorithm optimization steps: S1: Initialize the sparrow population, as shown in formula (1): In the formula, d represents the number of parameters to be optimized. In the present invention, the parameters to be optimized are the kernel parameter and the penalty factor. Therefore, d=2, and n is the number of sparrows in the population. In the present invention, the number of population initialization is set to 100. S2: Calculate the fitness of each individual in the sparrow population. In the present invention, the fitness is the accuracy of the test results of the test set; S3: According to the fitness of the sparrows in the group, the sparrows with higher fitness are the discoverers, and the rest are the followers; The finder is responsible for searching for food and providing directions for other sparrows to search for food; Followers mainly follow finders in their search for food; The sentinels are responsible for detecting predators and sending danger messages to the sparrow population, which then abandons food and migrates to safe areas. If the number of discoverers in the sparrow population is set to account for 20% of the total population, the sparrows will be sorted according to their individual fitness, and the sparrows with the top 20% fitness will be the discoverers; S4: Update the position of the discoverer in the sparrow population, which can be expressed as: Where t represents the number of iterations, i and j represent the subscript and superscript in formula (1), i represents the i-th sparrow, j represents the j-th dimension parameter of the i-th sparrow, itermax represents the maximum number of iterations, which can be expressed as T; Q is a random number that obeys the normal distribution; L is a matrix whose elements are all 1; R2 represents the safety value of the sparrow population position, and ST is the warning value of the sparrow population position; α is a uniform random number between 0 and 1, R2 is a uniform random number between 0 and 1, and ST is a constant between 0.5 and 1, which can be set during the initialization phase; When the safety value R2 is less than the warning value ST, it means that the position of the sparrow population is safe, and the search position of the discoverer in the population is larger; when the safety value R2 is less than the warning value ST, it means that the position of the sparrow population is unsafe, and the sparrow population should migrate; S5: Update the position of the follower in the sparrow population, which can be expressed as: In the formula, is the optimal position of the discoverer in the sparrow population at the current iteration, x worst is the worst position of an individual in the global sparrow population, A is a 1×d matrix whose elements are randomly selected from -1 or 1, A + It is expressed as: A + =A T (CHALLENGE ACCEPTED T ) -1 (4) S6: Update the position of the sentinel in the sparrow population, which can be expressed as: In the formula, represents the optimal position in the sparrow population at the tth iteration, β is the step size, and the value is a random number. The random number is normally distributed with a mean of 0 and a variance of 1; f i is the fitness of the current iteration number of the i-th sparrow, f g 、f w is the optimal fitness and the worst fitness before the t+1th iteration; ε is a very small constant to avoid f i =f w , which makes formula (5) meaningless; Formula (5) shows that when the fitness of the vigilant is equal to f g When , it indicates that the sparrow that has been warned is in the best position, and the sparrow will flee to the nearby area; If the fitness of the sentinel is not equal to f g , then the alerter escapes to the current optimal position; The swarm intelligence algorithm sparrow algorithm is used to optimize the kernel parameters and penalty factors in the support vector machine.

Images