Expressway sudden accident traffic flow dynamic prediction method based on neural network

Abstract

The invention relates to a highway sudden accident traffic flow dynamic prediction method based on a neural network, and the method comprises the steps: obtaining the traffic flow data of an accident-prone road section, and constructing a traffic flow parameter time sequence according to the traffic flow data; according to the traffic flow parameter time sequence, training an RBF neural network model to obtain a traffic flow prediction model; and predicting the traffic flow through the traffic flow prediction model. The method has the advantages that the original traffic flow parameters are preprocessed, the traffic flow is predicted through the RBF neural network model, the traffic state of the traffic flow can be accurately predicted, more useful information can be extracted from complex urban traffic data, and therefore more accurate and comprehensive data support is provided for traffic management and planning.

Description

Technical Field

[0001] The invention relates to the technical field of traffic flow prediction, and more specifically, to a method for dynamically predicting traffic flow of highway accidents based on neural networks. Background Art

[0002] There are two main reasons for traffic jams and accidents. One is that the overall situation of the traffic network cannot be grasped, and traffic induction and reasonable traffic allocation cannot be carried out effectively; the other is that travelers cannot know the real-time changes in road conditions and travel blindly. Therefore, accurate traffic flow prediction can not only provide a decision-making basis for the formulation of transportation organization plans by traffic management departments, make timely or advance responses to highway traffic congestion, and ensure the normal operation of highways, but also help travelers to make reasonable travel plans and reduce time and economic losses.

[0003] At present, short-term traffic flow prediction is mostly concentrated on urban roads, and there is little research on short-term traffic flow prediction on highways, especially short-term traffic flow prediction considering traffic accidents. Summary of the invention

[0004] The purpose of the present invention is to address the deficiencies of the prior art and to propose a method for dynamically predicting traffic flow in highway accidents based on a neural network.

[0005] First, a method for predicting the dynamic traffic flow of highway accidents based on neural networks is provided, including:

[0006] S1. Obtain traffic flow data of accident-prone sections and construct a traffic flow parameter time series based on the traffic flow data;

[0007] S2. According to the traffic flow parameter time series, the RBF neural network model is trained to obtain a traffic flow prediction model;

[0008] S3. Predict traffic flow through a traffic flow prediction model.

[0009] Preferably, in S1, the traffic flow data includes flow data, speed data and lane occupancy data.

[0010] Preferably, in S1, the traffic flow data is preprocessed, and a traffic flow parameter time series is constructed based on the preprocessed traffic flow data; the preprocessing is used to eliminate invalid data.

[0011] Preferably, S2 comprises:

[0012] S201, determining a delay step and a prediction step according to the traffic flow parameter time series, and constructing an input-output matrix;

[0013] S202. Divide the traffic flow parameter time series into training samples and test samples according to the input-output matrix;

[0014] S203. Train the RBF neural network model with the training samples to obtain a traffic flow prediction model.

[0015] Preferably, S2 further includes:

[0016] S204. Conduct a simulation test on the traffic flow prediction model with the test samples.

[0017] Preferably, in S202, the training samples and test samples are normalized.

[0018] In a second aspect, a highway sudden accident traffic flow dynamic prediction system based on a neural network is provided, which is used to execute any of the methods in the first aspect, including:

[0019] An acquisition module, configured to acquire traffic flow data of accident-prone sections and construct a traffic flow parameter time series according to the traffic flow data;

[0020] A training module, configured to train the RBF neural network model according to the traffic flow parameter time series to obtain a traffic flow prediction model;

[0021] A prediction module, configured to predict the traffic flow through the traffic flow prediction model.

[0022] In a third aspect, a computer storage medium is provided. The computer storage medium stores a computer program; when the computer program runs on a computer, the computer is enabled to execute any of the methods in the first aspect.

[0023] In a fourth aspect, an electronic device is provided, including:

[0024] A memory, configured to store the computer program;

[0025] A processor, configured to execute the computer program to implement any of the methods in the first aspect.

[0026] The beneficial effects of the present invention are as follows: By preprocessing the original traffic flow parameters and predicting the traffic flow through the RBF neural network model, the present invention can accurately predict the traffic state, extract more useful information from complex urban traffic data, and thus provide more accurate and comprehensive data support for traffic management and planning. Description of the Drawings

[0027] Figure 1 It is a flowchart of a highway sudden accident traffic flow dynamic prediction method provided by an embodiment of the present invention;

[0028] Figure 2 Flow chart of another traffic flow dynamic prediction method based on neural network provided by an embodiment of the present invention;

[0029] Figure 3 Flow chart of yet another traffic flow dynamic prediction method based on neural network provided by an embodiment of the present invention;

[0030] Figure 4 Schematic diagram of the time distribution of highway traffic accidents provided by an embodiment of the present invention;

[0031] Figure 5 Schematic diagram of the spatial distribution of highway traffic accidents provided by an embodiment of the present invention;

[0032] Figure 6 Schematic diagram of the prediction result of the RBF neural network model provided by an embodiment of the present invention;

[0033] Figure 7 Schematic diagram of the structure of the neural network and neurons provided by an embodiment of the present invention;

[0034] Figure 8 Schematic diagram of the relative error of the prediction result of the RBF neural network model provided by an embodiment of the present invention;

[0035] Figure 9 Schematic diagram of the structure of the device of the hardware operating environment provided by an embodiment of the present invention. Detailed implementation manners

[0036] The present invention will be further described below in conjunction with embodiments. The description of the following embodiments is only for helping to understand the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several modifications can still be made to the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

[0037] Embodiment 1:

[0038] To solve the problems of the prior art, Embodiment 1 of the present application provides a traffic flow dynamic prediction method for highway sudden accidents based on neural network. By locating accident-prone sections and using the method of neural network to construct time series of relevant traffic flow parameters, it can accurately predict the traffic state, extract more useful information from complex urban traffic data, and thus provide more accurate and comprehensive data support for traffic management and planning.

[0039] Specifically, as Figure 1 shown, the traffic flow dynamic prediction method for highway sudden accidents based on neural network includes:

[0040] S1. Obtain the traffic flow data of accident-prone sections, and construct a time series of traffic flow parameters based on the traffic flow data.

[0041] In S1, the traffic flow data includes flow data, speed data, and lane occupancy data. There are various ways to obtain the traffic flow data of accident-prone sections. In an optional implementation, construct a multi-source heterogeneous dataset of traffic accidents. As Figure 4 and Figure 5 shown, conduct traffic accident feature analysis from two dimensions of time (start time, end time, and duration) and space (nearest distance, farthest distance, and spacing), clarify accident-prone sections and accident-prone times, provide support for traffic management and traffic decision-making, and also provide directions for key traffic flow prediction objects. In another optional implementation, traffic flow data can also be collected at preset sections and times. For example, collect traffic flow data during peak traffic hours (such as the morning peak from 7:00 to 9:00 and the evening peak from 17:00 to 19:00) and at preset accident-prone sections (such as sharp curves, steep slopes, tunnel entrances, etc.), which can reduce the data collection cost and avoid redundant data collection throughout the day and across all sections.

[0042] In addition, in S1, it is necessary to preprocess the traffic flow data and construct a time series of traffic flow parameters based on the preprocessed traffic flow data; the preprocessing is used to eliminate invalid data. The elimination criteria for invalid data can be set artificially, such as obvious invalid data with a speed of 0, etc.

[0043] S2. Train the RBF neural network model according to the time series of traffic flow parameters to obtain a traffic flow prediction model.

[0044] As Figure 2 shown, S2 includes:

[0045] S201. Determine the delay step and prediction step according to the time series of traffic flow parameters, and construct an input-output matrix.

[0046] The delay step and prediction step can be determined artificially through data characteristics and autocorrelation analysis. For example, if matrix A = [1:2:7; 1:3:10], through preset steps (delay step or prediction step) (assumed to be 2 and 3), that is, establish a matrix with a step of 2 between the two end values of 1 - 7. Similarly, establish a matrix with a step of 3 between 1 - 10 in the second row. Then the input-output matrix can be expressed as A = [1 3 5 7; 1 4 7 10].

[0047] S202. Divide the time series of traffic flow parameters into training samples and test samples according to the input-output matrix.

[0048] In S202, the training samples and test samples are divided according to a time window. The first 80% of the data can be used as training samples, and the last 20% as test samples, ensuring that there is no time interval between the training samples and the test samples. The training samples are used to train and learn the model, and the test samples are used to judge the prediction accuracy of the model. At the same time, both the training samples and the test samples are normalized so that the data is limited within a certain range, thereby eliminating the adverse effects caused by singular sample data.

[0049] Specifically, the normalization calculation can be performed through the following formula:

[0050] ;

[0051] In the formula: is the data after normalization, is the variable the maximum value of, is the variable the minimum value of.

[0052] S203. Train the RBF neural network model with the training samples to obtain a traffic flow prediction model.

[0053] Among them, the RBF neural network consists of an input layer, a hidden layer, and an output layer. The neurons in the hidden layer are activated using radial basis functions. The Gaussian function is one of the most commonly used radial basis functions in the RBF neural network, and its mathematical expression is:

[0054]

[0055] Among them, represents the output of the Gaussian function, represents the distance between the input data and the center point, represents the standard deviation of the Gaussian function.

[0056] The role of the radial basis function in the RBF neural network is to determine the activation value of the hidden layer neurons according to the distance between the input data and the center point. The smaller the distance, the larger the activation value, and vice versa. This can achieve a non-linear mapping of the input data, enabling the RBF neural network to approximate complex non-linear functional relationships.

[0057] The neural network model in this case uses a multi-layer perceptron, and its basic architecture is constructed by stacking multiple single-layer units layer by layer. Each single-layer unit contains multiple independent neurons inside, and there is a rightward connection relationship between the neurons, and it only exists between adjacent two-layer perceptrons.

[0058] Such as Figure 7As shown in the figure, the structure of each neuron consists of two parts: a linear transformation unit and a non-linear activation function. When a neuron receives an input vector of any dimension, it converts it into a single scalar output, enabling a perceptron constructed by multiple neurons to map one vector to another. The calculation formula is as follows:

[0059]

[0060] Among them, is the input vector, is the weight matrix, is the bias vector, is the output vector, is the non-linear activation function.

[0061] The multi-layer perceptron is formed by arranging single-layer structures in series. The weights, biases, and activation functions from the input layer to the hidden layer are set as , and (·), respectively. Therefore, the calculation formula can be expressed as:

[0062]

[0063] S3. Predict the traffic flow through the traffic flow prediction model.

[0064] Example 2:

[0065] Based on Example 1, Example 2 of the present application provides a more specific dynamic traffic flow prediction method for highway sudden accidents based on neural networks, including:

[0066] S1. Obtain the traffic flow data of accident-prone sections and construct a time series of traffic flow parameters according to the traffic flow data;

[0067] S2. Train the RBF neural network model according to the time series of traffic flow parameters to obtain a traffic flow prediction model.

[0068] As Figure 3 shown, S2 includes:

[0069] S201. Determine the delay step and prediction step according to the time series of traffic flow parameters and construct an input-output matrix;

[0070] S202. Divide the time series of traffic flow parameters into training samples and test samples according to the input-output matrix;

[0071] S203. Train the RBF neural network model with the training samples to obtain a traffic flow prediction model.

[0072] S204. Perform a simulation test on the traffic flow prediction model using test samples.

[0073] In this embodiment, after continuously training the RBF neural network model and determining an appropriate radial basis function expansion speed, predictions are made on the test samples and the prediction results are output. The model is evaluated using the mean absolute error (MAE), root mean square error (RMSE), relative error (RE), and coefficient of determination R2.

[0074] The RBF neural network is a three-layer feedforward network that uses radial basis functions as the activation function for the hidden layer neurons. Essentially, however, it is a two-layer network where the input layer nodes only transmit the input signals to the hidden layer, the hidden layer nodes are composed of radial basis functions, and the output layer nodes are usually linear functions.

[0075] The activation function of the hidden layer nodes can be selected in different forms, and the commonly used one is still the Gaussian kernel function, that is:

[0076] ;

[0077] The output of the RBF neural network can be expressed as:

[0078] ;

[0079] The learning process of the RBF network is divided into two stages: In the first stage, the training samples are clustered. In the second stage, after determining the parameters of the hidden layer, according to the training samples, using the least squares principle, the weights w of the output layer are obtained. After completing the learning in the second stage, the parameters of the hidden layer and the output layer can be corrected simultaneously according to the sample signals to further improve the nonlinear mapping accuracy of the network.

[0080] Specifically, this application first determines the distance between classes , take equal to the average minimum distance between the training samples multiplied by , that is:

[0081]

[0082] Among them, is the scaling factor, and can be modified through it. It is controlled by to determine the number of clusters generated.

[0083] Since too many or too few cluster numbers will both reduce the accuracy, considering the influence of the convergence accuracy and convergence speed, generally choose = 1.5.

[0084] When is determined, judge the distance between the input sample and the cluster center ​ :

[0085]

[0086] And determine whether the sample is included in the cluster. When , recalculate :

[0087]

[0088] Among them, is the number of samples in the th class.

[0089] For example, , then a new class is created. In the newly created class, its center is determined by the given sample, and then repeated calculations are performed until all classes are found.

[0090] Through the above algorithm, this application only needs to traverse all training samples once to complete clustering, which is very effective for constructing the hidden layer of the RBF neural network and has better performance compared with K-means clustering and SOFM clustering.

[0091] Specifically, the calculation formula for the absolute mean error is:

[0092] ;

[0093] The calculation formula for the root mean square error is:

[0094] ;

[0095] The calculation formula for the relative error is:

[0096] ;

[0097] The calculation formula for the coefficient of determination R2 is:

[0098] ;

[0099] In the formula: N is the number of data, is the true value, is the predicted value, and y is the average of the true values.

[0100] S3. Predict the traffic flow through the traffic flow prediction model.

[0101] The following is only used to explain this application through actual cases and does not limit this application.

[0102] The traffic accident data and traffic flow data used in this application are from the northbound section of I-880N in the United States, with a total length of 46 miles (about 74 kilometers). There are 206 detectors, with a relatively high detector coverage rate and relatively complete data.

[0103] First, please refer to Figure 4 and Figure 5 By statistically analyzing the occurrence times of 456 traffic accidents on the northbound section of I-880N in May 2021, it is found that the occurrence times of traffic accidents are concentrated from 12:00 to 17:00, and accidents are most likely to occur at 12:00, accounting for about 9.2%. The road section is divided into 46 segments at a unit of 1 kilometer, and by statistically analyzing the number of accidents occurring in each segment, it is found that accidents mostly occur in the middle section of the road, that is, at the 23-28 kilometer section, and the accident occurrence proportion is about 23.7%.

[0104] Through the above analysis of traffic accident characteristics, it is found that the 23-28 kilometer section of the road is an accident-prone section. In this study, 27 kilometers is taken as the research object, and the all-day traffic flow from May 1st to May 7th is collected and statistically analyzed in units of 5 minutes to construct a training sample traffic flow time series; the all-day traffic flow from May 29th to May 31st of this section is collected and statistically analyzed in units of 5 minutes to construct a test sample traffic flow time series.

[0105] The traffic flow in the training sample and the test sample is normalized, and the processing result is limited between 0 and 1, so as to eliminate the adverse effects caused by singular sample data.

[0106] Subsequently, a model is established using MATLAB, and the constructor function provided by the toolbox is called:

[0107]

[0108] Among them, P is the training sample input, T is the training sample output, spread is the radial basis function expansion speed, and the model is trained using the training sample data. After the model training is completed, the test sample is input into the trained network, and the prediction result is output. The function:

[0109]

[0110] Among them, P is the test sample input, T is the test sample output, and net is the trained network. To better compare the test results, the prediction result is de-normalized.

[0111] Please refer to Table 1, Figure 6 and Figure 8, according to the model output results, the absolute mean error and root mean square error of the model prediction are small, and the model prediction results are good. The R2 is 0.9888, indicating a high model fitting degree. At the same time, by comparing the traffic flows at two accident points in the test set, the prediction relative error of accident 1 is 7.64%, and the prediction relative error of accident 2 is 9.80%, both within the acceptable range, indicating that the model can also well predict the traffic flow during the accident period.

[0112] Table 1

[0113]

[0114] In summary, the technical solution provided by this application can effectively adapt to complex traffic systems and environments. Especially when dealing with traffic flow prediction problems under highway traffic accident conditions, it demonstrates excellent performance.

[0115] By locating accident-prone sections and using the neural network method to construct time series of relevant traffic flow parameters, it can accurately predict the traffic state, extract more useful information from complex urban traffic data, and thus provide more accurate and comprehensive data support for traffic management and planning.

[0116] It improves the accuracy of short-term traffic flow prediction, provides important technical support for the development of intelligent transportation systems, helps traffic managers conduct traffic regulation more effectively, reduces traffic congestion, and improves the transportation efficiency of roads.

[0117] It should be noted that the same or similar parts in this embodiment and Embodiment 1 can be referred to each other and will not be elaborated in this application.

[0118] Embodiment 3:

[0119] Based on Embodiment 2, Embodiment 3 of this application provides a dynamic traffic flow prediction system for highway sudden accidents based on neural networks, including:

[0120] An acquisition module, configured to acquire traffic flow data of accident-prone sections and construct a time series of traffic flow parameters according to the traffic flow data;

[0121] A training module, configured to train an RBF neural network model according to the time series of traffic flow parameters to obtain a traffic flow prediction model;

[0122] A prediction module, configured to predict the traffic flow through the traffic flow prediction model.

[0123] Specifically, the system provided in this embodiment is the system corresponding to the methods provided in Embodiments 1 and 2. Therefore, the same or similar parts in this embodiment and Embodiments 1 and 2 can be referred to each other and will not be elaborated in this application.

[0124] Example 4:

[0125] Based on Examples 1 - 3, as Figure 8 shown, Example 4 of this application provides a terminal structure for the hardware operating environment.

[0126] The terminal in the embodiment of the present invention can be a PC, or a smart phone, a tablet computer, an e - book reader, an MP3 (Moving Picture Experts Group Audio Layer III) player, an MP4 (Moving Picture Experts Group Audio Layer IV) player, a portable computer and other movable terminal devices with display functions.

[0127] As Figure 8 shown, the terminal may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard). Optionally, the user interface 1003 may further include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI - FI interface). The memory 1005 may be a high - speed RAM memory or a stable memory (non - volatile memory), such as a disk memory. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0128] Those skilled in the art can understand that Figure 9 the terminal structure shown in

[0129] does not limit the terminal, and may include more or fewer components than shown in the figure, or combine some components, or have different component arrangements. Figure 9 As

[0130] In Figure 9 the terminal shown, the network interface 1004 is mainly used to connect to the background server and communicate with the background server for data; the user interface 1003 is mainly used to connect to the client (user side) and communicate with the client for data; and the processor 1001 can be used to call the traffic flow dynamic prediction program stored in the memory 1005 and perform the following operations:

[0131] Obtain traffic flow data on accident-prone sections and construct a time series of traffic flow parameters based on the traffic flow data;

[0132] According to the traffic flow parameter time series, the RBF neural network model is trained to obtain a traffic flow prediction model;

[0133] Traffic flow is predicted through traffic flow prediction model.

[0134] Furthermore, the processor 1001 may call the traffic flow dynamic prediction program stored in the memory 1005, and further perform the following operations:

[0135] Based on the analysis of the spatiotemporal characteristics of traffic accidents, a multi-source heterogeneous data set of traffic accidents is constructed. At the same time, traffic flow data of accident-prone sections are obtained based on the analysis of the spatiotemporal characteristics of traffic accidents.

[0136] Furthermore, the processor 1001 may call the traffic flow dynamic prediction program stored in the memory 1005, and further perform the following operations:

[0137] The traffic flow data is preprocessed, and a traffic flow parameter time series is constructed based on the preprocessed traffic flow data; the preprocessing is used to eliminate invalid data.

[0138] Furthermore, the processor 1001 may call the traffic flow dynamic prediction program stored in the memory 1005, and further perform the following operations:

[0139] According to the traffic flow parameter time series, determining the delay step and the prediction step, and constructing an input-output matrix;

[0140] According to the input-output matrix, the traffic flow parameter time series is divided into training samples and test samples;

[0141] The RBF neural network model is trained through training samples to obtain the traffic flow prediction model.

[0142] The traffic flow prediction model is simulated and tested through test samples.

[0143] Furthermore, the processor 1001 may call the traffic flow dynamic prediction program stored in the memory 1005, and further perform the following operations:

[0144] The training samples and the test samples are normalized.

[0145] Specifically, the terminal structure provided in this embodiment is the terminal structure corresponding to the method provided in Embodiment 2. Therefore, the same or similar parts in this embodiment as those in Embodiment 2 can be referred to each other and will not be described in detail in this application.

Claims

1. A method for predicting traffic flow dynamics of highway accidents based on neural networks, characterized in that: The following steps are involved: S1. Obtain traffic flow data of accident-prone sections and construct a traffic flow parameter time series based on the traffic flow data; S2. According to the traffic flow parameter time series, the RBF neural network model is trained to obtain a traffic flow prediction model; S3. Predict traffic flow through a traffic flow prediction model.

2. The method for predicting traffic flow dynamics of highway accidents based on neural networks according to claim 1 is characterized in that: In S1, the traffic flow data includes flow data, speed data and lane occupancy data.

3. The method for predicting traffic flow dynamics of highway accidents based on neural networks according to claim 2 is characterized in that: In S1, the traffic flow data is preprocessed, and a traffic flow parameter time series is constructed based on the preprocessed traffic flow data; the preprocessing is used to eliminate invalid data.

4. The method for predicting traffic flow dynamics of highway accidents based on neural networks according to claim 3 is characterized in that S2 include: S201, determining a delay step and a prediction step according to the traffic flow parameter time series, and constructing an input-output matrix; S202, dividing the traffic flow parameter time series into training samples and test samples according to the input-output matrix; S203, training the RBF neural network model through training samples to obtain a traffic flow prediction model.

5. The method for predicting traffic flow dynamics of highway accidents based on neural network according to claim 4 is characterized in that: S2 also includes: S204: Perform simulation test on the traffic flow prediction model through test samples.

6. The method for predicting traffic flow dynamics of highway accidents based on neural networks according to claim 5 is characterized in that: In S202, the training samples and the test samples are normalized.

7. A dynamic prediction system for highway accident traffic flow based on neural network, characterized by: Used to perform the method according to any one of claims 1 to 6, comprising: An acquisition module is used to acquire traffic flow data of accident-prone sections and construct a traffic flow parameter time series based on the traffic flow data; A training module, used for training the RBF neural network model according to the traffic flow parameter time series to obtain a traffic flow prediction model; The prediction module is used to predict traffic flow through a traffic flow prediction model.

8. A computer storage medium, characterized in that The computer storage medium stores a computer program; when the computer program is executed on a computer, the computer executes any one of the methods described in claims 1 to 6.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the method according to any one of claims 1 to 6.

Images