A multi-level traffic accident prediction and correction system, method and product

Abstract

The embodiment of the present invention provides a multi-level traffic accident prediction and correction system, method and product, which relates to the field of computer technology. The system includes: a traffic accident identifier, which is used to divide the traffic accident log into four different groups to identify the target traffic accident record; a traffic accident predictor, which is used to determine the weights of multiple sensors of the target road section, and weight the traffic data collected by multiple sensors of the target road section in the current period according to the determined weights, and use the weighted traffic data as the candidate state, and use the traffic data collected by multiple sensors of the target road section during the normal traffic period as the healthy state, and determine the health coefficient of the target road section based on the distribution difference between the candidate state and the healthy state; a traffic accident corrector, which is used to use a convolutional neural network as a classifier to verify the authenticity of the traffic accident occurring on the target road section, and obtain the final prediction result of the target road section, so as to improve the accuracy of traffic accident prediction for the road.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of computer technology, and in particular to a multi-level traffic accident prediction and correction system, method and product. Background Art

[0002] In the current era of rapid development of automobiles, the occurrence of traffic accidents will seriously affect people's work, life and even life. In order to avoid the occurrence of traffic accidents, it is particularly important to predict traffic accidents on roads, so as to take relevant defense measures for roads in advance based on the prediction results of road traffic accidents, avoid the occurrence of traffic accidents and ensure road traffic safety. However, the prediction accuracy of traffic accident prediction on roads in the current related technology is low. Therefore, how to improve the prediction accuracy of traffic accident prediction on roads is a technical problem to be solved by the present invention. Summary of the invention

[0003] Based on the above technical problems, the embodiments of the present invention provide a multi-level traffic accident prediction and correction system, method and product, aiming to improve the prediction accuracy of traffic accident prediction for roads.

[0004] A first aspect of an embodiment of the present invention provides a multi-level traffic accident prediction and correction system, the system comprising:

[0005] A traffic accident identifier, used for dividing the traffic accident log into four different groups to identify target traffic accident records, wherein the target traffic accident records are records of traffic accidents with an occurrence frequency less than a preset frequency and an average traffic accident recovery time less than a preset time;

[0006] A traffic accident predictor is used to determine the weights of each of the multiple sensors of the target road section, and weight the traffic data collected by the multiple sensors of the target road section in the current time period according to the determined weights, and use the weighted traffic data as the candidate state, and use the traffic data collected by the multiple sensors of the target road section in the normal traffic period as the healthy state, determine the distribution difference between the candidate state and the healthy state, and determine the health coefficient of the target road section according to the distribution difference, wherein the health coefficient represents the traffic condition of the target road section in the next time period;

[0007] A traffic accident corrector, used to compare the health coefficient of the target road section in the next time period with a health threshold, and when the health coefficient of the next time period is higher than the health threshold, use a convolutional neural network as a classifier to verify the authenticity of the traffic accident occurring in the target road section, so as to obtain a final prediction result of the target road section, wherein the final prediction result represents the corrected traffic condition of the target road section in the next time period;

[0008] Among them, the traffic accident predictor, the traffic accident corrector and the weights of each of the multiple sensors are obtained by training with the target traffic accident record as a training sample; the traffic data includes at least one of the following: vehicle speed, vehicle acceleration, vehicle temperature, whether the vehicle's automatic driving function is turned on, the vehicle's steering wheel angle, the value of the vehicle's limited slip differential, the number of vehicles on the road, the average speed of vehicles on the road, and the maximum speed on the road.

[0009] A second aspect of an embodiment of the present invention provides a multi-level traffic accident prediction and correction method, which is applied to the multi-level traffic accident prediction and correction system provided in the first aspect of an embodiment of the present invention, and the method at least includes:

[0010] The traffic accident log is divided into four different groups by a traffic accident identifier to identify target traffic accident records, wherein the target traffic accident records are records of traffic accidents whose occurrence frequency is less than a preset frequency and whose average traffic accident recovery time is less than a preset time;

[0011] Determine the weights of the multiple sensors of the target road section through a traffic accident predictor, and weight the traffic data collected by the multiple sensors of the target road section in the current time period according to the determined weights, and use the weighted traffic data as the candidate state, and use the traffic data collected by the multiple sensors of the target road section in the normal traffic period as the healthy state, determine the distribution difference between the candidate state and the healthy state, and determine the health coefficient of the target road section according to the distribution difference, wherein the health coefficient represents the traffic condition of the target road section in the next time period;

[0012] By means of a traffic accident corrector, the health coefficient of the target road section in the next time period is compared with a health threshold value. When the health coefficient of the next time period is higher than the health threshold value, a convolutional neural network is used as a classifier to verify the authenticity of the traffic accident occurring in the target road section, so as to obtain a final prediction result of the target road section, wherein the final prediction result represents the corrected traffic condition of the target road section in the next time period;

[0013] Among them, the traffic accident predictor, the traffic accident corrector and the weights of each of the multiple sensors are obtained by training with the target traffic accident record as a training sample; the traffic data includes at least one of the following: vehicle speed, vehicle acceleration, vehicle temperature, whether the vehicle's automatic driving function is turned on, the vehicle's steering wheel angle, the value of the vehicle's limited slip differential, the number of vehicles on the road, the average speed of vehicles on the road, and the maximum speed on the road.

[0014] A third aspect of an embodiment of the present invention provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the multi-level traffic accident prediction and correction method of the second aspect of the embodiment of the present invention is implemented.

[0015] A fourth aspect of an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the multi-level traffic accident prediction and correction method of the second aspect of an embodiment of the present invention is implemented.

[0016] In the multi-level traffic accident prediction and correction system of the embodiment of the present invention, the target traffic accident record is obtained by screening through the traffic accident identifier to train the traffic accident predictor, the traffic accident corrector and the weights of each of the multiple sensors; the traffic accident predictor calculates the distribution difference between the traffic data of the current time period and the traffic data of the normal traffic period for the target road section, so as to predict the health coefficient of the target road section in the next time period according to the distribution difference, and obtain the traffic condition of the target road section in the next time period; the predicted health coefficient is corrected by the traffic accident corrector to identify effective traffic accident warnings to further improve the accuracy of traffic accident prediction and improve the prediction accuracy of traffic accident prediction for roads. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0018] Figure 1 is a structural block diagram of a multi-level traffic accident prediction and correction system shown in one embodiment of the present invention;

[0019] Figure 2 It is a four-quadrant schematic diagram based on a traffic accident log proposed in an embodiment of the present invention;

[0020] Figure 3 It is a schematic diagram of the structure and parameters of a convolutional neural network in a traffic accident corrector proposed in one embodiment of the present invention;

[0021] Figure 4 is a health coefficient curve diagram of a multi-level traffic accident prediction and correction system shown in one embodiment of the present invention;

[0022] Figure 5It is a structural schematic diagram of a multi-level traffic accident prediction and correction system according to an embodiment of the present invention;

[0023] Figure 6 is a flowchart of the steps of a multi-level traffic accident prediction and correction method provided by an embodiment of the present invention;

[0024] Figure 7 It is a schematic diagram of an electronic device shown in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] Please refer to Figure 1 , Figure 1 FIG. 1 is a structural block diagram of a multi-level traffic accident prediction and correction system according to an embodiment of the present invention. Figure 1 As shown, the multi-level traffic accident prediction and correction system of this embodiment may include at least: a traffic accident identifier, a traffic accident predictor and a traffic accident corrector.

[0027] The traffic accident identifier is used to divide the traffic accident log into four different groups to identify target traffic accident records, wherein the target traffic accident records are records of traffic accidents whose occurrence frequency is less than a preset frequency and whose average traffic accident recovery time is less than a preset time.

[0028] In this embodiment, the historical traffic accident logs can be analyzed by the traffic accident identifier, and the traffic accident logs can be divided into four different groups to identify one group from the four different groups as the target traffic accident record. The target traffic accident record can be used as a training sample to train the traffic accident predictor, the traffic accident corrector and the respective weights of multiple sensors.

[0029] Among them, the target traffic accident record is a record of traffic accidents whose occurrence frequency is less than the preset frequency and whose average recovery time is less than the preset time. The target traffic accident record may include road-related corresponding traffic data, such as multiple sensor data related to the road. The traffic data in this embodiment includes at least one of the following: vehicle speed, vehicle acceleration, vehicle temperature, whether the vehicle automatic driving function is turned on, vehicle steering wheel angle, vehicle limited slip differential value, number of road vehicles, average speed of road vehicles, and maximum speed of the road. Among them, vehicle speed, vehicle acceleration, vehicle temperature, whether the vehicle automatic driving function is turned on, vehicle steering wheel angle, and vehicle limited slip differential (Limited Slip Differential, LSD) value are relevant data of the vehicle itself on the road; the number of road vehicles, average speed of road vehicles, and maximum speed of the road are relevant data of the road.

[0030] A traffic accident predictor is used to determine the respective weights of multiple sensors of a target road section, and weight the traffic data collected by the multiple sensors of the target road section in a current time period according to the determined weights, and use the weighted traffic data as a candidate state, and use the traffic data collected by the multiple sensors of the target road section in a normal traffic period as a healthy state, determine the distribution difference between the candidate state and the healthy state, and determine the health coefficient of the target road section based on the distribution difference, wherein the health coefficient represents the traffic condition of the target road section in the next time period.

[0031] In this embodiment, the trained traffic accident predictor can predict the health coefficient of the target road section to determine whether a traffic accident will occur in the target road section, wherein the target road section is the road section to be predicted for a traffic accident.

[0032] Specifically, the traffic accident predictor can first determine the weights of the multiple sensors of the target road section, and according to the determined weights of the multiple sensors of the target road section, weight the traffic data collected by the multiple sensors of the target road section in the current period to obtain weighted traffic data (i.e., the weighted traffic data collected by the multiple sensors in the current period); then, the weighted traffic data is used as the candidate state, and the traffic data collected by the multiple sensors of the target road section during the normal traffic period is used as the healthy state, and the distribution difference between the candidate state and the healthy state is determined. For example, several indicators such as Kullback-Leibler (KL) divergence can be used to quantify the distribution difference. After obtaining the distribution difference, the health coefficient of the target road section in the next period can be determined according to the distribution difference. Among them, the health coefficient represents the traffic condition of the target road section in the next period.

[0033] A traffic accident corrector is used to compare the health coefficient of the target road section in the next time period with a health threshold. When the health coefficient of the next time period is higher than the health threshold, a convolutional neural network is used as a classifier to verify the authenticity of the traffic accident occurring on the target road section, so as to obtain a final prediction result of the target road section, and the final prediction result represents the traffic condition of the target road section after correction in the next time period.

[0034] In this embodiment, the traffic accident predictor may have a high false alarm rate. In order to avoid inaccurate traffic accident prediction, this embodiment proposes a traffic accident corrector that can correct the prediction result to improve reliability. After the traffic accident predictor predicts the health coefficient of the target road section in the next period, the health coefficient of the target road section in the next period can be input into the traffic accident corrector.

[0035] The traffic accident corrector first compares the health coefficient of the target road section in the next period with the health threshold. When the health coefficient of the target road section in the next period is higher than the health threshold, the traffic accident corrector can use a convolutional neural network as a classifier to process the health coefficient and verify the authenticity of the traffic accident on the target road section to obtain the final prediction result of the target road section. The final prediction result represents the corrected traffic condition of the target road section in the next period. The traffic condition of this embodiment can include at least one of the following: traffic accident, no traffic accident, and traffic congestion.

[0036] In this embodiment, a health threshold is set in advance, which is the maximum value of the health coefficient representing traffic safety. The specific value of the health threshold can be freely set according to needs, such as the health threshold is 0.2. Furthermore, a corresponding health threshold can be set for each road, or all roads can correspond to the same health threshold, which is not limited in this embodiment.

[0037] In this embodiment, target traffic accident records are obtained by screening through a traffic accident identifier to train the traffic accident predictor, the traffic accident corrector and the weights of each of the multiple sensors; the traffic accident predictor calculates the distribution difference between the traffic data of the current time period and the traffic data of the normal traffic period for the target road section to predict the health coefficient of the target road section in the next time period based on the distribution difference, and obtain the traffic conditions of the target road section in the next time period; the predicted health coefficient is corrected by the traffic accident corrector to identify effective traffic accident warnings to further improve the accuracy of traffic accident predictions and improve the prediction accuracy of traffic accident predictions for roads.

[0038] In combination with the above embodiments, in one implementation, the present invention further provides a multi-level traffic accident prediction and correction system, in which the step of "dividing the traffic accident log into four different groups to identify the target traffic accident record" performed by the above traffic accident identifier may specifically include steps S11 to S12:

[0039] Step S11: using a real traffic data set including the traffic accident log as a data source, and based on a four-quadrant graph, analyzing the correlation between the frequency of traffic accidents and the average time to recover from the traffic accidents.

[0040] In this embodiment, the traffic accident identifier uses a real traffic data set as a data source, and analyzes the data source based on a four-quadrant diagram to analyze the correlation between the frequency of traffic accidents in the data source and the average time to recover from traffic accidents. The real traffic data set of this embodiment includes a traffic accident log. It can be understood that the real traffic data set of this embodiment includes both traffic accident logs and normal traffic data.

[0041] In this embodiment, based on the frequency of traffic accidents and the average time of traffic accident recovery, a preset frequency corresponding to the frequency of traffic accidents and a preset time corresponding to the average time of traffic accident recovery can be set respectively; then, based on the preset frequency, the preset time and the real traffic data set, a four-quadrant diagram between the frequency of traffic accidents and the average time of traffic accident recovery is established. In the four-quadrant diagram, the real traffic data set is divided into four quadrants according to the preset frequency and the preset time.

[0042] Among them, the first quadrant is a subset of real traffic data where the frequency of traffic accidents is higher than the preset frequency and the average time for traffic accident recovery is less than the preset time, that is, the first quadrant contains a subset of real traffic data with high frequency but can be recovered in a short time. The recommendation for this type of real traffic is to perform traffic maintenance regularly. The second quadrant is a subset of real traffic data where the frequency of traffic accidents is higher than the preset frequency and the average time for traffic accident recovery is higher than the preset time, that is, the second quadrant contains a subset of high-risk and serious real traffic data with high frequency but a long time to recover. Generally, this type of traffic accident data has no reference value and will not play any role in accident prediction.

[0043] The third quadrant is a subset of real traffic data where the frequency of traffic accidents is lower than the preset frequency and the average time for traffic accident recovery is less than the preset time, that is, the third quadrant contains a subset of real traffic data with low frequency and short recovery time. The recommendation for this type of real traffic is to use a passive maintenance strategy for traffic maintenance. It is understandable that the real traffic data subset in the third quadrant also includes normal traffic data where the frequency of traffic accidents is zero and the average time for traffic accident recovery is zero. This normal traffic data is the traffic data collected by the sensor during normal traffic hours.

[0044] The fourth quadrant is a subset of real traffic data in which the frequency of traffic accidents is lower than the preset frequency and the average time for traffic accident recovery is higher than the preset time. That is, the fourth quadrant contains a subset of real traffic data with low frequency and long time to recover. The suggestions for this type of real traffic are the same as those in the second quadrant and are not suitable for accident prediction.

[0045] Step S12: Based on the association relationship, the subset of real traffic data in the third quadrant is determined as the target traffic accident record.

[0046] In this embodiment, based on the correlation between the frequency of traffic accidents and the average time to recover from traffic accidents, the subset of real traffic data in the third quadrant obtained by dividing based on the correlation can be determined as the target traffic accident record. It can be understood that the target traffic accident record in this embodiment includes not only traffic accident logs but also normal traffic data.

[0047] like Figure 2 As shown, Figure 2 FIG. 1 is a four-quadrant schematic diagram based on a traffic accident log proposed in an embodiment of the present invention. Figure 2 In the figure, the horizontal axis is the average time for traffic accident recovery, in seconds; the vertical axis is the frequency of traffic accidents, in years. Figure 2 The preset frequency corresponding to the frequency of occurrence of traffic accidents and the preset time corresponding to the average time for traffic accident recovery are graphically divided into four quadrants. Among them, no traffic accidents in the upper right quadrant are expected, because any problems in this area may be identified and solved in traffic diversion. The upper left quadrant contains traffic accidents with high frequency but short recovery time. The recommendation for such traffic accidents is to perform traffic maintenance regularly. The frequency of traffic accidents in the lower left quadrant is low and the average recovery time is short; in this case, a passive maintenance strategy is recommended because it will be the most economical strategy. The most critical traffic accidents occur in the lower right quadrant. These occasional traffic accidents usually result from gradual degradation that can be revealed through data analysis and are not regarded as representative of predictions. Therefore, the traffic accident records obtained from the lower left quadrant, i.e., the third quadrant, can be used as training data for this embodiment.

[0048] In combination with the above embodiments, in one implementation, the present invention further provides a multi-level traffic accident prediction and correction system. In this embodiment, the traffic accident predictor and the traffic accident corrector can be trained through the following steps:

[0049] Step S21: marking a traffic accident label on the traffic data collected by multiple sensors on the sample road section in the target traffic accident record during multiple historical abnormal traffic periods.

[0050] In this embodiment, the target traffic accident records screened out by the traffic accident identifier as training samples include: sample traffic data collected by multiple sensors on the sample road section during multiple historical abnormal traffic periods, and sample traffic data collected by multiple sensors on the sample road section during historical normal traffic periods.

[0051] In this embodiment, the traffic accident label can be marked on the traffic data collected by multiple sensors of the sample road section in the target traffic accident record during multiple historical abnormal traffic periods. That is to say, for multiple historical abnormal traffic periods, the label "traffic accident occurred" is marked on the traffic data collected by multiple sensors of the sample road section during each historical abnormal traffic period. The sample road section in this embodiment is a road section used as a training sample.

[0052] Step S22: Determine the sample weights of the multiple sensors of the sample road section through the traffic accident predictor to be trained, and weight the sample traffic data collected by the multiple sensors of the sample road section in the target traffic accident record during the current historical abnormal traffic period according to the determined sample weights, and use the weighted sample traffic data as the sample candidate state, and use the sample traffic data collected by the multiple sensors of the sample road section in the target traffic accident record during the historical normal traffic period as the sample health state, determine the sample distribution difference between the sample candidate state and the sample health state, and determine the sample health coefficient of the sample road section in the next historical abnormal traffic period according to the sample distribution difference.

[0053] The traffic accident predictor to be trained in this embodiment is a basic model that needs to be trained as a traffic accident predictor. The sample weights corresponding to each of the multiple sensors of the sample road section can be determined by the traffic accident predictor to be trained. The sample weights are the weights corresponding to the sensors during the training process. Then, the traffic accident predictor to be trained can weight the sample traffic data collected by the multiple sensors of the sample road section in the target traffic accident record during the current historical abnormal traffic period according to the sample weights corresponding to each of the multiple sensors of the determined sample road section, and obtain the weighted sample traffic data of the multiple sensors of the sample road section during the current historical abnormal traffic period. Among them, the current historical abnormal traffic period is any one of the multiple historical abnormal traffic periods, and the next historical abnormal traffic period is the next period of the current historical abnormal traffic period among the multiple historical abnormal traffic periods. The sample traffic data is the traffic data during the training process.

[0054] In this embodiment, the traffic accident predictor to be trained can determine the weighted sample traffic data of multiple sensors of the sample road section during the current historical abnormal traffic period as the weighted sample traffic data, and use the weighted sample traffic data as the sample candidate state, and the sample candidate state is the candidate state in the training process. Then, the traffic accident predictor to be trained can use the sample traffic data collected by multiple sensors of the sample road section in the target traffic accident record during the historical normal traffic period as the sample health state, and the sample health state is the health state in the training process. Among them, the historical normal traffic period represents the historical period when no traffic accidents occurred, and the historical abnormal traffic period represents the period when traffic accidents occurred.

[0055] In this embodiment, after the trained traffic accident predictor determines the sample candidate state and the sample health state, the sample distribution difference between the sample candidate state and the sample health state can be determined, and the sample health coefficient of the sample road section in the next historical abnormal traffic period can be determined based on the sample distribution difference. The sample distribution difference is the distribution difference during the training process, and the sample health coefficient is the health coefficient during the training process.

[0056] Step S23: Through the traffic accident corrector to be trained, the sample health coefficient of the sample road section in the next historical abnormal traffic period is compared with the health threshold. When the sample health coefficient is higher than the health threshold, the convolutional neural network to be trained is used as a classifier to verify the authenticity of the traffic accident occurring on the sample road section to obtain the sample prediction result of the sample road section.

[0057] In this embodiment, after the sample health coefficient of the sample road section in the next historical abnormal traffic period is predicted by the traffic accident predictor to be trained, the sample health coefficient of the sample road section in the next historical abnormal traffic period can be input into the traffic accident corrector to be trained, and the traffic accident corrector to be trained compares the sample health coefficient of the sample road section in the next historical abnormal traffic period with a preset health threshold, and when the sample health coefficient is higher than the health threshold, the traffic accident corrector to be trained can use the convolutional neural network to be trained as a classifier to process the sample health coefficient to verify the authenticity of the traffic accident occurring in the sample road section in the next historical abnormal traffic period, so as to obtain a sample prediction result of the sample road section, which characterizes the predicted traffic condition of the sample road section in the next historical abnormal traffic period after correction.

[0058] Step S24: Based on the sample prediction result and the traffic accident label corresponding to the next historical abnormal traffic period, the convolutional neural network to be trained in the traffic accident predictor to be trained and the traffic accident corrector to be trained is trained to obtain the traffic accident predictor and the traffic accident corrector.

[0059] In this embodiment, the traffic accident corrector to be trained can be based on the sample prediction results and the traffic accident label corresponding to the sample road section in the next historical abnormal traffic period, to train the traffic accident predictor to be trained and the convolutional neural network to be trained in the traffic accident corrector to be trained until the training is completed to obtain the trained traffic accident predictor and the trained convolutional neural network to be trained, thereby obtaining the traffic accident predictor and the traffic accident corrector.

[0060] In combination with the above embodiments, the present invention further provides a multi-level traffic accident prediction and correction system, in which the "determining the sample weights of the multiple sensors of the sample road section" in the above step S22 may specifically include steps S31 and S32:

[0061] Step S31: From the time dimension, perform global average pooling processing on the sample traffic data collected by each sensor in the historical period to obtain the channel description factor z of each sensor.

[0062] In this embodiment, the attention mechanism is used to evaluate and emphasize the relative importance of different sensor data channels, recognizing that different sensor data channels have different effects on the state of road traffic. Specifically, the traffic accident predictor to be trained can perform global average pooling processing on the sample traffic data collected by each sensor of the sample road in historical time periods (including historical abnormal traffic periods and historical normal traffic periods) from the time dimension to obtain a lightweight channel description factor for each sensor. .

[0063] .

[0064] In this embodiment, for each sensor of the sample road, the sample weight w corresponding to each sensor can be determined according to the above formula based on the channel description factor z of each sensor. and is a trainable parameter of the traffic accident predictor to be trained, r is a scaling factor, N is the length of the time series, ReLU is a ReLU function, and Sigmoid is a Sigmoid function. In an optional example, r=16.

[0065] In combination with the above embodiments, in one implementation, the present invention further provides a multi-level traffic accident prediction and correction system. In this embodiment, the traffic accident predictor performs "taking weighted traffic data as a candidate state, taking traffic data collected by multiple sensors of the target road section during a normal traffic period as a healthy state, and determining the distribution difference between the candidate state and the healthy state" which can specifically include step S41:

[0066] Step S41: Determine the distribution difference by the following formula: :

[0067] .

[0068] In this embodiment, the MMD algorithm can be used to compare the data distribution based on the reproducing kernel Hilbert space, providing a non-parametric alternative method for evaluating the distribution difference. Specifically, the distribution difference between the candidate state and the healthy state is determined by the above MMD formula , to empirically estimate the distance between p and q defined by MMD. Where H is the reproducing kernel Hilbert space, f is the mapping function, let X = and Y = Represent traffic data with distributions of p and q, respectively. p reflects the healthy state and q reflects the candidate state. Each x n and n Corresponding to the traffic data of the time series with a length of L collected from the nth sensor, N represents the total number of the multiple sensors. In one example, N = 31, and the time window L = 1500.

[0069] In getting the distribution difference Afterwards, we can analyze the distribution difference Normalization is performed to determine the health coefficient of the target road section.

[0070] Furthermore, in one embodiment, when training the traffic accident predictor to be trained, the traffic accident predictor to be trained can be trained by the above-mentioned MMD formula, where p reflects the sample health state and q reflects the sample candidate state; since MMD essentially lacks trainable parameters, based on this, this embodiment combines the attention mechanism through the channel description factor z to convert the trainable parameters and A traffic accident predictor to be trained is introduced, thereby allowing the use of target traffic accident records as a source of training data.

[0071] In combination with the above embodiments, in one implementation, the present invention further provides a multi-level traffic accident prediction and correction system. In this embodiment, the step of "using a convolutional neural network as a classifier to verify the authenticity of a traffic accident occurring on a target road section to obtain a final prediction result of the target road section" performed by the traffic accident corrector may specifically include steps S51 to S55:

[0072] Step S51: After the health coefficient of the next time period is input into the convolutional neural network, 6 first feature maps of size 24×24 are obtained in the first convolutional layer.

[0073] In this embodiment, the health coefficient is a health indicator curve graph with a dimensional feature of N×L, where N represents the total number of multiple sensors, and L represents the number of time units of the sensor in the time dimension; the traffic accident corrector of this embodiment is a traffic accident corrector based on a convolutional neural network CNN, and the convolutional neural network at least includes: a first convolutional layer, a subsampling layer, a second convolutional layer, a first fully connected layer and a second fully connected layer, and each layer contains trainable parameters.

[0074] When the traffic accident corrector in this embodiment determines that the health coefficient of the target road section in the next time period is higher than the health threshold, the health coefficient of the next time period can be input into the convolutional neural network in the traffic accident corrector, and 6 first feature maps of size 24×24 are obtained in the first convolutional layer in the convolutional neural network; the first feature map is the feature map obtained by processing the first convolutional layer.

[0075] That is to say, the first convolutional layer in this embodiment includes 6 first feature maps, each of which has a size of 24×24, and each unit in these first feature maps is connected to a 127×8 convolution kernel from the output layer of the convolutional neural network.

[0076] Step S52: In the subsampling layer, the six first feature maps are processed by a 2×2 average pooling operator to obtain six second feature maps with a size of 12×12.

[0077] In this embodiment, after obtaining the 6 first feature maps output by the first convolutional layer, the 6 first feature maps can be input into the subsampling layer, in which the 6 first feature maps are processed by a 2×2 average pooling operator to obtain 6 second feature maps of size 12×12, which are feature maps obtained by the subsampling layer. When the average pooling process is performed in the subsampling layer and the moving pooling frame is not overlapped with the previous layer (i.e., the first convolutional layer).

[0078] Step S53: In the second convolution layer, a convolution kernel of size 2×2 is used to process the six second feature maps to obtain 32 third feature maps of size 6×6.

[0079] In this embodiment, the six second feature maps obtained by the subsampling layer are input into the second convolutional layer. In the second convolutional layer, the six second feature maps are processed by a convolution kernel of size 2×2 to obtain 32 third feature maps of size 6×6. The third feature map is the feature map obtained by the second convolutional layer.

[0080] Step S54: Input the 32 third feature maps into the first fully connected layer to obtain 128 first neurons.

[0081] In this embodiment, after the second convolutional layer outputs 32 third feature maps, the 32 third feature maps can be input into the first fully connected layer to obtain the output of the first fully connected layer: 128 first neurons. The first neuron is a neuron processed by the first fully connected layer. Each neuron is connected to a 6×6 neighborhood (i.e., a convolution kernel of size 6×6) through all 32 third feature maps of the second convolutional layer.

[0082] Step S55: inputting 128 of the first neurons into the second fully connected layer to obtain one second neuron, and processing the second neuron through a sigmoid activation function to obtain a final prediction result of the target road section.

[0083] In this embodiment, after the first fully connected layer obtains 128 first neurons, the 128 first neurons can be input into the second fully connected layer to obtain the output of the second fully connected layer: 1 second neuron, where the second neuron is the neuron obtained by processing the second fully connected layer. Then, the second neuron is processed by the sigmoid activation function to obtain a binary output: the final prediction result of the target road section. The final prediction result of the target road section can indicate whether it represents a true warning or a false warning.

[0084] The traffic accident corrector using the convolutional neural network structure proposed in this embodiment can effectively classify and correct erroneous traffic accident predictions, thereby improving the accuracy and reliability of the traffic accident prediction system.

[0085] like Figure 3 As shown, Figure 3 FIG. 1 is a schematic diagram of the structure and parameters of a convolutional neural network in a traffic accident corrector proposed in one embodiment of the present invention. Figure 3 In the example, the health coefficient is the input matrix of the convolutional neural network, which includes two dimensions: time and features. The time dimension spans 150 time units (equivalent to 3000 seconds), and the feature dimension includes 31 different sensors. Therefore, the input matrix generated by the health coefficient in this embodiment is . And, in Figure 3In the convolutional neural network, there are 5 layers (excluding the input layer), namely, C1 layer (first convolution layer), S2 layer (sub-collection layer), C3 layer (second convolution layer), F4 layer (first fully connected layer) and F5 layer (second fully connected layer). The feature maps processed by the C1 layer are 6 first feature maps of size 24×24; the feature maps processed by the S2 layer are 6 second feature maps of size 12×12; the feature maps processed by the C3 layer are 32 third feature maps of size 6×6; the neurons processed by the F4 layer are 128 first neurons; and the neurons processed by the F5 layer are 1 second neuron. Finally, the second neuron is processed by the sigmoid activation function to obtain the final output (i.e., the final prediction result) representing the true warning or false warning. In one example, the ReLU function is used as the activation function after each convolution step. After two rounds of convolution, the health coefficient is converted into a 128-length vector, which is then compressed through a fully connected network. Finally, a sigmoid function is used to perform binary classification to evaluate the effectiveness of the fault alarm.

[0086] In combination with the above embodiments, in one implementation, the present invention also provides a multi-level traffic accident prediction and correction system. In the system, the traffic accident corrector is also used to output the warning information of the target road section and the traffic accident evacuation strategy for the target road section when the health coefficient of the target road section exceeds the health threshold in multiple consecutive time periods and the final prediction result of the target road section in multiple consecutive time periods is that a traffic accident occurs.

[0087] In this embodiment, when the traffic accident corrector determines that the health coefficient of the target road section exceeds the health threshold in multiple consecutive time periods, and the final prediction results of the target road section in multiple consecutive time periods are all traffic accidents, the traffic accident corrector can output warning information for the target road section and a traffic accident evacuation strategy for the target road section, wherein the warning information is used to prompt relevant users to pay attention and take corresponding preventive measures.

[0088] In one embodiment, if Figure 4 As shown, Figure 4 : is a health coefficient curve diagram of a multi-level traffic accident prediction and correction system shown in an embodiment of the present invention. Figure 4 The above curve is a health coefficient curve of a road in one year. Since the variability of the health coefficient hinders the accurate assessment of the road monitoring health status, this embodiment uses a sliding time window to smooth the health coefficient curve to obtain Figure 4 The following graph. Figure 4The curve chart below shows that the higher the health coefficient, the worse the health status of road traffic. If the health coefficient exceeds the health threshold and remains above the threshold for more than one month, it can be considered that the road needs traffic maintenance and the corresponding traffic accident evacuation strategy measures are taken. The dotted box indicates that the health coefficient of the road exceeds the threshold and lasts for one month, at which time the road traffic needs to be maintained. The solid box represents the maintenance of road traffic during this period. After this, the health coefficient of the road will be significantly reduced. And, after this maintenance event, the health coefficient is significantly reduced, which represents the ideal road traffic health status.

[0089] In another embodiment, if Figure 5 As shown, Figure 5 FIG. 1 is a schematic diagram of a multi-level traffic accident prediction and correction system according to an embodiment of the present invention. Figure 5 In the present invention, a multi-level traffic accident prediction and correction system includes: (a) a traffic accident discriminator, (b) a traffic accident predictor, and (c) a traffic accident corrector. First, in the traffic accident discriminator, the real traffic data set containing the traffic accident log is processed, and the target traffic accident record is obtained by analyzing the fault analyzer. Then, the traffic accident predictor and the traffic accident corrector are trained with the target traffic accident record. Specifically, the traffic accident predictor divides the target traffic accident record into a healthy state and a candidate state, and then the attention mechanism-MMD algorithm is used to determine the distribution difference between the healthy state and the candidate state to obtain the health coefficient. Finally, the health coefficient is input into the traffic accident corrector, and the traffic accident corrector first compares the health coefficient with the health threshold. When it is determined that the health coefficient is higher than the health threshold, the health coefficient is input into the convolutional neural network in the traffic accident corrector to obtain the final prediction result output by the convolutional neural network, so as to carry out the corresponding road maintenance strategy based on the prediction result.

[0090] Based on the same inventive concept, an embodiment of the present invention provides a multi-level traffic accident prediction and correction method, which is applied to the multi-level traffic accident prediction and correction system described in any of the above embodiments. Figure 6 , Figure 6 FIG. 1 is a flowchart of a multi-level traffic accident prediction and correction method provided by an embodiment of the present invention. Figure 6 As shown, the multi-level traffic accident prediction and correction method of this embodiment includes the following steps:

[0091] Step S61: using a traffic accident identifier, the traffic accident log is divided into four different groups to identify target traffic accident records, wherein the target traffic accident records are records of traffic accidents with an occurrence frequency less than a preset frequency and an average traffic accident recovery time less than a preset time;

[0092] Step S62: Determine the weights of the multiple sensors of the target road section through the traffic accident predictor, and weight the traffic data collected by the multiple sensors of the target road section in the current time period according to the determined weights, and use the weighted traffic data as the candidate state, and use the traffic data collected by the multiple sensors of the target road section in the normal traffic period as the healthy state, determine the distribution difference between the candidate state and the healthy state, and determine the health coefficient of the target road section according to the distribution difference, and the health coefficient represents the traffic condition of the target road section in the next time period;

[0093] Step S63: using a traffic accident corrector, comparing the health coefficient of the target road section in the next time period with a health threshold, and when the health coefficient of the next time period is higher than the health threshold, using a convolutional neural network as a classifier to verify the authenticity of the traffic accident occurring in the target road section, so as to obtain a final prediction result of the target road section, wherein the final prediction result represents the corrected traffic condition of the target road section in the next time period;

[0094] Among them, the traffic accident predictor, the traffic accident corrector and the weights of each of the multiple sensors are obtained by training with the target traffic accident record as a training sample; the traffic data includes at least one of the following: vehicle speed, vehicle acceleration, vehicle temperature, whether the vehicle's automatic driving function is turned on, the vehicle's steering wheel angle, the value of the vehicle's limited slip differential, the number of vehicles on the road, the average speed of vehicles on the road, and the maximum speed on the road.

[0095] Optionally, the traffic accident logs are divided into four different groups to identify target traffic accident records, including:

[0096] Taking a real traffic data set including the traffic accident log as a data source, and based on a four-quadrant graph, analyzing the correlation between the frequency of traffic accidents and the average time to recover from the traffic accidents;

[0097] Based on the association relationship, the real traffic data subset in the third quadrant is determined as the target traffic accident record.

[0098] Optionally, the method further comprises the following steps for training a traffic accident predictor and a traffic accident corrector:

[0099] Marking traffic accident labels on traffic data collected by multiple sensors on a sample road section in the target traffic accident record during multiple historical abnormal traffic periods;

[0100] Determine the sample weights of the multiple sensors of the sample road section through the traffic accident predictor to be trained, and weight the sample traffic data collected by the multiple sensors of the sample road section in the target traffic accident record during the current historical abnormal traffic period according to the determined sample weights, and use the weighted sample traffic data as the sample candidate state, and use the sample traffic data collected by the multiple sensors of the sample road section in the target traffic accident record during the historical normal traffic period as the sample health state, determine the sample distribution difference between the sample candidate state and the sample health state, and determine the sample health coefficient of the sample road section in the next historical abnormal traffic period according to the sample distribution difference;

[0101] By using the to-be-trained traffic accident corrector, the sample health coefficient of the sample road section in the next historical abnormal traffic period is compared with the health threshold value. When the sample health coefficient is higher than the health threshold value, the to-be-trained convolutional neural network is used as a classifier to verify the authenticity of the traffic accident occurring in the sample road section, so as to obtain the sample prediction result of the sample road section;

[0102] Based on the sample prediction results and the traffic accident label corresponding to the next historical abnormal traffic period, the convolutional neural network to be trained in the traffic accident predictor to be trained and the traffic accident corrector to be trained is trained to obtain the traffic accident predictor and the traffic accident corrector.

[0103] Optionally, the determining of the sample weights of the plurality of sensors of the sample road section includes:

[0104] From the time dimension, the sample traffic data collected by each sensor in the historical period is globally averaged and pooled to obtain the channel description factor z of each sensor;

[0105] Based on the channel description factor z of each sensor, the sample weight corresponding to each sensor is determined according to the following formula:

[0106] ;

[0107] in, and are the trainable parameters of the traffic accident predictor to be trained, r is the scaling factor, N is the length of the time series, ReLU is the ReLU function, and Sigmoid is the Sigmoid function.

[0108] Optionally, the weighted traffic data is used as the candidate state, the traffic data collected by multiple sensors of the target road section during a normal traffic period is used as the healthy state, and the distribution difference between the candidate state and the healthy state is determined, including:

[0109] The distribution difference is determined by the following formula :

[0110] ;

[0111] In which, let X = and Y = Respectively represent traffic data with distributions of p and q, where p reflects the health state and q reflects the candidate state. Each x n and n Corresponding to the traffic data of the time series with a length of L collected from the nth sensor, N represents the total number of multiple sensors, H is the reproduction kernel Hilbert space, and f is the mapping function.

[0112] Optionally, the health coefficient is a health index curve graph with a dimension feature of N×L, where N represents the total number of multiple sensors, and L represents the number of time units of the sensor in the time dimension; the convolutional neural network includes at least: a first convolutional layer, a subsampling layer, a second convolutional layer, a first fully connected layer, and a second fully connected layer;

[0113] The use of a convolutional neural network as a classifier to verify the authenticity of a traffic accident occurring on a target road section to obtain a final prediction result of the target road section includes:

[0114] After inputting the health coefficient of the next period into the convolutional neural network, six first feature maps of a size of 24×24 are obtained in the first convolutional layer;

[0115] In the subsampling layer, the six first feature maps are processed by a 2×2 average pooling operator to obtain six second feature maps with a size of 12×12;

[0116] In the second convolution layer, a convolution kernel of size 2×2 is used to process the six second feature maps to obtain 32 third feature maps of size 6×6;

[0117] Inputting the 32 third feature maps into the first fully connected layer to obtain 128 first neurons;

[0118] Input 128 of the first neurons into the second fully connected layer to obtain 1 second neuron, and process the second neuron through a sigmoid activation function to obtain a final prediction result of the target road section.

[0119] Optionally, the method further comprises,

[0120] Through the traffic accident corrector, when the health coefficient of the target section exceeds the health threshold in multiple consecutive time periods and the final prediction result of the target section in multiple consecutive time periods is that a traffic accident occurs, the warning information of the target section and the traffic accident evacuation strategy for the target section are output.

[0121] It should be noted that, for the sake of simplicity, the method embodiments are described as a series of action combinations, but those skilled in the art should be aware that the embodiments of the present invention are not limited by the order of the actions described, because according to the embodiments of the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

[0122] Based on the same inventive concept, another embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps in the multi-level traffic accident prediction and correction method as described in any of the above embodiments of the present invention are implemented.

[0123] Based on the same inventive concept, another embodiment of the present invention provides an electronic device 700, such as Figure 7 shown. Figure 7 7 is a schematic diagram of an electronic device according to an embodiment of the present invention. The electronic device includes a memory 702, a processor 701, and a computer program stored in the memory and executable on the processor, and when executed by the processor, the steps in the multi-level traffic accident prediction and correction method according to any of the above embodiments of the present invention are implemented.

[0124] As for the method embodiment, since it is basically similar to the system embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the system embodiment.

[0125] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0126] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, devices, or computer program products. Therefore, embodiments of the present invention may take the form of complete hardware embodiments, complete software embodiments, or embodiments combining software and hardware. Furthermore, embodiments of the present invention may take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0127] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0128] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0129] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0130] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention.

[0131] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or terminal device including the elements.

[0132] The above is a detailed introduction to a multi-level traffic accident prediction and correction system, method and product provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A multi-level traffic accident prediction and correction system, characterized in that: The system at least comprises: A traffic accident identifier, used to classify the traffic accident log into four different groups to identify target traffic accident records, wherein the target traffic accident records are records of traffic accidents with an occurrence frequency less than a preset frequency and an average traffic accident recovery time less than a preset time; A traffic accident predictor is used to determine the weights of each of the multiple sensors of the target road section, and weight the traffic data collected by the multiple sensors of the target road section in the current time period according to the determined weights, and use the weighted traffic data as the candidate state, and use the traffic data collected by the multiple sensors of the target road section in the normal traffic period as the healthy state, determine the distribution difference between the candidate state and the healthy state, and determine the health coefficient of the target road section according to the distribution difference, wherein the health coefficient represents the traffic condition of the target road section in the next time period; A traffic accident corrector, used to compare the health coefficient of the target road section in the next time period with a health threshold, and when the health coefficient of the next time period is higher than the health threshold, use a convolutional neural network as a classifier to verify the authenticity of the traffic accident occurring in the target road section, so as to obtain a final prediction result of the target road section, wherein the final prediction result represents the corrected traffic condition of the target road section in the next time period; The traffic accident predictor, the traffic accident corrector and the weights of each of the multiple sensors are obtained by training with the target traffic accident record as a training sample; the traffic data includes at least one of the following: vehicle speed, vehicle acceleration, vehicle temperature, whether the vehicle automatic driving function is turned on, vehicle steering wheel angle, vehicle limited slip differential value, number of road vehicles, average speed of road vehicles, and maximum speed of road; The method of using the weighted traffic data as the candidate state and the traffic data collected by multiple sensors of the target road section during a normal traffic period as the healthy state, and determining the distribution difference between the candidate state and the healthy state includes: The distribution difference is determined by the following formula : ； In which, let X = and Y = Respectively represent traffic data with distributions of p and q, where p reflects the health state and q reflects the candidate state. Each x n and n Corresponding to the traffic data of the time series with a length of L collected from the nth sensor, N represents the total number of multiple sensors, H is the reproduction kernel Hilbert space, and f is the mapping function.

2. The system according to claim 1, characterized in that The traffic accident logs are divided into four different groups to identify target traffic accident records, including: Taking a real traffic data set including the traffic accident log as a data source, and based on a four-quadrant graph, analyzing the correlation between the frequency of traffic accidents and the average time to recover from the traffic accidents; Based on the association relationship, the real traffic data subset in the third quadrant is determined as the target traffic accident record.

3. The system according to claim 2, characterized in that The traffic accident predictor and the traffic accident corrector are trained by the following steps: Marking traffic accident labels on traffic data collected by multiple sensors on a sample road section in the target traffic accident record during multiple historical abnormal traffic periods; Determine the sample weights of the multiple sensors of the sample road section through the traffic accident predictor to be trained, and weight the sample traffic data collected by the multiple sensors of the sample road section in the target traffic accident record during the current historical abnormal traffic period according to the determined sample weights, and use the weighted sample traffic data as the sample candidate state, and use the sample traffic data collected by the multiple sensors of the sample road section in the target traffic accident record during the historical normal traffic period as the sample health state, determine the sample distribution difference between the sample candidate state and the sample health state, and determine the sample health coefficient of the sample road section in the next historical abnormal traffic period according to the sample distribution difference; By using the to-be-trained traffic accident corrector, the sample health coefficient of the sample road section in the next historical abnormal traffic period is compared with the health threshold value. When the sample health coefficient is higher than the health threshold value, the to-be-trained convolutional neural network is used as a classifier to verify the authenticity of the traffic accident occurring in the sample road section, so as to obtain the sample prediction result of the sample road section; Based on the sample prediction results and the traffic accident label corresponding to the next historical abnormal traffic period, the convolutional neural network to be trained in the traffic accident predictor to be trained and the traffic accident corrector to be trained is trained to obtain the traffic accident predictor and the traffic accident corrector.

4. The system according to claim 3, characterized in that The determining of the sample weights of the plurality of sensors of the sample road section comprises: From the time dimension, the sample traffic data collected by each sensor in the historical period is globally averaged and pooled to obtain the channel description factor z of each sensor; Based on the channel description factor z of each sensor, the sample weight corresponding to each sensor is determined according to the following formula: ； in, and are the trainable parameters of the traffic accident predictor to be trained, r is the scaling factor, N is the length of the time series, ReLU is the ReLU function, and Sigmoid is the Sigmoid function.

5. The system according to claim 1, characterized in that The health coefficient is a health index curve graph with a dimension feature of N×L, where N represents the total number of multiple sensors, and L represents the number of time units of the sensor in the time dimension; The convolutional neural network at least includes: a first convolutional layer, a subsampling layer, a second convolutional layer, a first fully connected layer, and a second fully connected layer; The use of a convolutional neural network as a classifier to verify the authenticity of a traffic accident occurring on a target road section to obtain a final prediction result of the target road section includes: After the health coefficient of the next period is input into the convolutional neural network, six first feature maps of a size of 24×24 are obtained in the first convolutional layer; In the subsampling layer, the six first feature maps are processed by a 2×2 average pooling operator to obtain six second feature maps with a size of 12×12; In the second convolution layer, a convolution kernel of size 2×2 is used to process the six second feature maps to obtain 32 third feature maps of size 6×6; Inputting the 32 third feature maps into the first fully connected layer to obtain 128 first neurons; Input 128 of the first neurons into the second fully connected layer to obtain 1 second neuron, and process the second neuron through a sigmoid activation function to obtain a final prediction result of the target road section.

6. The system according to any one of claims 1 to 5, characterized in that: The traffic accident corrector is also used to output the warning information of the target road section and the traffic accident evacuation strategy for the target road section when the health coefficient of the target road section in multiple consecutive time periods exceeds the health threshold and the final prediction result of the target road section in multiple consecutive time periods is that a traffic accident has occurred.

7. A multi-level traffic accident prediction and correction method, characterized in that: The multi-level traffic accident prediction and correction system applied to any one of claims 1 to 6, the method at least comprising: The traffic accident log is divided into four different groups by a traffic accident identifier to identify target traffic accident records, wherein the target traffic accident records are records of traffic accidents whose occurrence frequency is less than a preset frequency and whose average traffic accident recovery time is less than a preset time; Determine the weights of the multiple sensors of the target road section through a traffic accident predictor, and weight the traffic data collected by the multiple sensors of the target road section in the current time period according to the determined weights, and use the weighted traffic data as the candidate state, and use the traffic data collected by the multiple sensors of the target road section in the normal traffic period as the healthy state, determine the distribution difference between the candidate state and the healthy state, and determine the health coefficient of the target road section according to the distribution difference, wherein the health coefficient represents the traffic condition of the target road section in the next time period; By means of a traffic accident corrector, the health coefficient of the target road section in the next time period is compared with a health threshold value. When the health coefficient of the next time period is higher than the health threshold value, a convolutional neural network is used as a classifier to verify the authenticity of the traffic accident occurring in the target road section, so as to obtain a final prediction result of the target road section, wherein the final prediction result represents the corrected traffic condition of the target road section in the next time period; The traffic accident predictor, the traffic accident corrector and the weights of each of the multiple sensors are obtained by training with the target traffic accident record as a training sample; the traffic data includes at least one of the following: vehicle speed, vehicle acceleration, vehicle temperature, whether the vehicle automatic driving function is turned on, vehicle steering wheel angle, vehicle limited slip differential value, number of road vehicles, average speed of road vehicles, and maximum speed of road; The method of using the weighted traffic data as the candidate state and the traffic data collected by multiple sensors of the target road section during a normal traffic period as the healthy state, and determining the distribution difference between the candidate state and the healthy state includes: The distribution difference is determined by the following formula : ； In which, let X = and Y = Respectively represent traffic data with distributions of p and q, where p reflects the health state and q reflects the candidate state. Each x n and n Corresponding to the traffic data of the time series with a length of L collected from the nth sensor, N represents the total number of multiple sensors, H is the reproduction kernel Hilbert space, and f is the mapping function.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the computer program is executed by the processor, the multi-level traffic accident prediction and correction method according to claim 7 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the multi-level traffic accident prediction and correction method as claimed in claim 7 is implemented.

Images