A congestion prediction method for high-risk accident areas

Abstract

The application relates to the technical field of data prediction, and provides a traffic jam prediction method for an accident high-risk area, which comprises the following steps: extracting the connection relationship between different detection points, the width of a road and the traffic flow of the detection points, determining the connection characteristic value of the detection points, and establishing the traffic flow vector and the traffic flow matrix of the detection points; establishing a connection characteristic matrix, a selectable path matrix and a traffic flow similarity matrix, and determining the fusion traffic characteristic matrix at a collection time; using a sine function to encode the collection time, obtaining the time point encoding value of the collection time, and obtaining traffic flow prediction data; and according to whether the collection time is within the rush hours of a workday, the traffic flow prediction data and the width of the road where the detection point is located, realizing traffic jam prediction. The application can improve the accuracy of traffic jam prediction.

Description

Technical Field

[0001] This application relates to the field of data prediction technology, specifically to a method for predicting congestion in high-risk accident areas. Background Technology

[0002] Urban traffic congestion prediction can help citizens avoid congested areas and improve their travel experience. It also provides traffic management departments with accurate data support, facilitating the development of short-term traffic management strategies. Furthermore, it can assist planning departments in the rational planning of road construction. Urban traffic congestion prediction not only alleviates current traffic pressure but is also a key measure to promote sustainable urban development and improve residents' quality of life.

[0003] Generally, traffic congestion can be predicted based on road traffic flow data. However, since different roads are interconnected, existing methods ignore the spatiotemporal relationships between interconnected roads, often leading to inaccurate traffic congestion predictions. Summary of the Invention

[0004] This application provides a congestion prediction method for high-risk accident areas to address the problem that traffic congestion prediction ignores the spatiotemporal correlation between interconnected roads, leading to inaccurate prediction results. The specific technical solution adopted is as follows:

[0005] One embodiment of this application provides a method for predicting congestion in high-risk accident areas, the method comprising the following steps:

[0006] Extract the connectivity between different detection points, the width of the road, and the traffic flow of different detection points at different collection times. Number the detection points, determine the connection feature values ​​of different detection points based on the connectivity, and establish the traffic flow vector and traffic flow matrix of each detection point at each collection time based on the traffic flow.

[0007] Based on the numbering and connection feature values ​​of different detection points, a connection feature matrix is ​​established. Based on the connectivity between different detection points, an optional path matrix is ​​established. Based on the correlation between the traffic flow vectors of different detection points at the same collection time, a traffic flow similarity matrix at the same collection time is determined. Based on the connection feature matrix, the optional path matrix, and the traffic flow similarity matrix at each collection time, a fused traffic feature matrix at each collection time is determined.

[0008] Each collection time is encoded using a sine function to obtain the time point encoding value of each collection time. Combined with the traffic flow matrix and the fused traffic feature matrix at the collection time, the traffic flow prediction data of each detection point at the next collection time after the collection time is obtained.

[0009] Traffic congestion can be predicted based on whether the data collection time falls within the weekday rush hour, traffic flow forecast data, and the width of the road where the monitoring point is located.

[0010] Furthermore, the method for determining the connection feature value of the detection point is as follows:

[0011] When two detection points are directly connected by a road and there are no other detection points between them, the connection feature value of the two detection points is assigned to 1.

[0012] When two detection points are connected by a road and there are other detection points between them, the sum of the number of other detection points in the road connecting the two detection points and 1 is used as the connection feature value of the two detection points.

[0013] Furthermore, the specific method for establishing the traffic flow vector and traffic flow matrix for each detection point at each collection time based on traffic flow is as follows:

[0014] Record any collection time as the target collection time. Arrange the traffic flow data collected at the same detection point at the target collection time and at each time interval before the target collection time into a column vector according to the time sequence, and obtain the traffic flow vector of the detection point at the target collection time.

[0015] Arrange the traffic flow vectors of all detection points from left to right in ascending order of their numbers to obtain the traffic flow matrix of the detection points at the target collection time.

[0016] Furthermore, the connection feature matrix is ​​specifically as follows:

[0017] The number of rows and columns in the connection feature matrix is ​​equal to the number of detection points. The nth row in the connection feature matrix... Line 1 Column data For the number The detection points and their numbers are The connection feature values ​​of the detection points, , , Indicates the number of detection points;

[0018] when When that happens, the corresponding connection feature value is assigned the value 1.

[0019] Furthermore, the optional path matrix is ​​specifically as follows:

[0020] The number of rows and columns in the optional path matrix is ​​equal to the number of detection points. Line 1 Column data For the number The detection point arrived at the numbered The normalized value of the number of different paths at the detection point;

[0021] when When that happens, the number of different paths will be assigned a value of 0.

[0022] Furthermore, the traffic flow similarity matrix at the same collection time is specifically as follows:

[0023] The number of rows and columns in the traffic flow similarity matrix at the same collection time is equal to the number of detection points. The number of rows and columns in the traffic flow similarity matrix is ​​the same as the number of detection points. Line 1 Column data For the number The detection points and their numbers are The correlation coefficient of the traffic flow vector at the same acquisition time for the detection points.

[0024] Furthermore, the formula for calculating the fused traffic feature matrix is:

[0025]

[0026] in, Indicates the first A fused traffic feature matrix for each data collection time; , and These represent the first preset parameter, the second preset parameter, and the third preset parameter, respectively, and the sum of the first preset parameter, the second preset parameter, and the third preset parameter is 1. Represents the connection feature matrix; Represents the optional path matrix; Indicates the first Traffic flow similarity matrix at each data collection time; This indicates the calculation of the Hadamard product.

[0027] Furthermore, the formula for calculating the time point encoding value at the acquisition time is:

[0028]

[0029] In the formula, Indicates the first The time point encoding value of each acquisition moment; This indicates the number of data collection moments within one hour; This represents the sine function.

[0030] Furthermore, the specific method for obtaining traffic flow prediction data for each detection point at the next collection time after the collection time by combining the traffic flow matrix at the collection time and the fused traffic feature matrix includes:

[0031] Update each traffic flow in the traffic flow matrix at the time of collection to the traffic flow and the corresponding time point code value at the time of collection, with the traffic flow and the time point code value separated by a comma, and obtain the updated traffic flow matrix at the time of collection.

[0032] Using the updated traffic flow matrix at the time of collection as the data matrix and the fused traffic feature matrix at the time of collection as the weight matrix, the traffic flow prediction data of each detection point at the next collection time after the current collection time is obtained according to the STGCN deep learning framework.

[0033] Furthermore, the method for predicting traffic congestion based on whether the data collection time falls within the weekday rush hour, traffic flow prediction data, and the width of the road where the detection point is located includes the following specific methods:

[0034] The time period of the data collection time is weighted according to whether the data collection time falls within the peak commuting hours on a weekday.

[0035] The ratio of the time period weight of the data collection time to the width of the road where the detection point is located is recorded as the first ratio of the detection point at the data collection time. The positive correlation between the first ratio of the detection point at the data collection time and the traffic flow prediction data at the next data collection time after the data collection time is recorded as the traffic congestion index of the detection point at the data collection time.

[0036] When the traffic congestion index is greater than the preset traffic congestion threshold, the detection point corresponding to the collection time of the traffic congestion index is marked as congested; when the traffic congestion index is less than or equal to the preset traffic congestion threshold, the detection point corresponding to the collection time of the traffic congestion index is marked as normal traffic.

[0037] The beneficial effects of this application are:

[0038] This application considers the intricate and intersecting nature of urban road systems, where the spatiotemporal relationships between interconnected roads significantly impact traffic congestion. To more clearly describe the spatiotemporal relationships between roads corresponding to different detection points, this application extracts road connectivity relationships between detection points, constructs a connectivity feature matrix, extracts the actual spatial connectivity relationships between each detection point, and constructs an optional path matrix. To capture the spatial correlation between different detection points, a traffic flow similarity matrix is ​​extracted at each data collection time. Furthermore, feature fusion is performed on the connectivity feature matrix, optional path matrix, and traffic flow similarity matrix to extract traffic information containing road connectivity relationships, spatial connectivity relationships, and spatial correlations between detection points for the road to be predicted as congested. The fused traffic feature matrix is ​​then determined for each data collection time. Based on the periodic variation of traffic flow on urban highways with natural days and weeks, a sine function is used to encode each collection time, obtaining the time-point encoding value for each collection time. This is then combined with the traffic flow matrix and fused traffic feature matrix at each collection time to obtain the traffic flow prediction data for the next collection time after the current collection time. This traffic flow prediction data represents the predicted traffic flow value for future collection times. Finally, based on whether the collection time falls within weekday rush hours, the traffic flow prediction data, and the width of the road where the collection point is located, traffic congestion prediction is achieved. This addresses the problem of inaccurate predictions due to neglecting the spatiotemporal relationships between interconnected roads, thus improving the accuracy of traffic congestion prediction. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 A schematic flowchart of a congestion prediction method for high-risk accident areas provided in one embodiment of this application;

[0041] Figure 2 This is a flowchart illustrating the process of obtaining a fused traffic feature matrix according to an embodiment of this application. Detailed Implementation

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

[0043] Please see Figure 1 The diagram illustrates a flowchart of a congestion prediction method for high-risk accident areas according to an embodiment of this application. The method includes the following steps:

[0044] Step S001: Extract the connectivity between different detection points, the width of the road, and the traffic flow of different detection points at different collection times. Number the detection points, determine the connection feature values ​​of different detection points based on the connectivity, and establish the traffic flow vector and traffic flow matrix of each detection point at each collection time based on the traffic flow.

[0045] By identifying detection points on roads where congestion is to be predicted, the number of vehicles passing through each detection point within the time interval between each data collection moment and the previous adjacent data collection moment is extracted and recorded as the traffic flow at that detection point at that time. The width of the road, the connectivity features of different detection points, and the different paths from one detection point to another are identified using an online map. These different paths from one detection point to another represent the connectivity relationships between the different detection points.

[0046] Preferably, as an embodiment of this application, when two detection points are directly connected by a road and there are no other detection points between the two detection points, the connection feature value of the two detection points is assigned to 1; when two detection points are connected by a road and there are other detection points between the two detection points, the sum of the number of other detection points contained in the road connecting the two detection points and 1 is used as the connection feature value of the two detection points.

[0047] It is understandable that there may be multiple paths connecting two detection points, so there exists a minimum number of other detection points contained in the path connecting the two detection points. In reality, the more detection points there are between two detection points, the more possible paths there are, and the more difficult the prediction becomes. In subsequent traffic prediction, the prediction model should pay more attention to the cases where other detection points exist, so that the prediction results are more consistent with reality. In the process of calculating the connection feature value between two detection points, adding the number of other detection points to 1 is to avoid the inability to distinguish between the two connection cases when only one other detection point is included.

[0048] In this embodiment, the time interval is set to 5 minutes, and traffic flow data from detection points within 2 hours prior to each analyzed collection time is collected. Based on the analyzed collection time and the traffic flow data from the detection points within 2 hours prior to that time, road congestion prediction is performed for the analyzed collection time. In practical applications, as other implementation methods, implementers can determine the sampling frequency and the number of samples according to actual conditions; this application does not impose any special restrictions.

[0049] Different detection points are numbered starting from 1 using natural numbers. Any collection time is recorded as the target collection time. Traffic flow data collected at the same detection point at the target collection time and at each time interval before the target collection time are arranged into a column vector in chronological order to obtain the traffic flow vector of the detection point at the target collection time. The traffic flow vectors of all detection points are arranged from left to right in ascending order of the detection point number to obtain the traffic flow matrix of the detection point at the target collection time.

[0050] It is understood that in this embodiment, the time interval is set to 5 minutes, and the traffic flow of the detection points is collected within 2 hours before the target collection time. That is, the traffic flow vector of each detection point contains 24 traffic flow data.

[0051] The same method can be used to obtain the traffic flow vector and traffic flow matrix of any detection point at any collection time.

[0052] At this point, the traffic flow vector and traffic flow matrix of each detection point at each collection time are obtained.

[0053] Step S002: Based on the numbering and connection feature values ​​of different detection points, establish a connection feature matrix; based on the connectivity between different detection points, establish an optional path matrix; based on the correlation between traffic flow vectors of different detection points at the same collection time, determine the traffic flow similarity matrix at the same collection time; based on the connection feature matrix, optional path matrix, and traffic flow similarity matrix at each collection time, determine the fused traffic feature matrix for each collection time.

[0054] In urban road systems, roads are intricate and intersecting, and the spatiotemporal relationships between interconnected roads have a significant impact on traffic congestion. Therefore, in order to more clearly describe the spatiotemporal relationships between roads corresponding to different detection points, a connection feature matrix, an optional path matrix, and a traffic flow similarity matrix at each collection time are constructed.

[0055] First, the road network is complex, and distant detection points may not be directly connected, while nearby detection points may be directly connected. Traditional connectivity analysis methods that directly analyze the location of detection points cannot accurately represent the road connectivity between them. Therefore, a connectivity feature matrix is ​​established based on the connectivity feature values ​​of different detection points.

[0056] In this context, the number of rows and columns in the connection feature matrix is ​​equal to the number of detection points, and the number of rows and columns in the connection feature matrix is ​​the nth row. Line 1 Column data For the number The detection points and their numbers are The connection feature values ​​of the detection points, , , This indicates the number of testing points.

[0057] It is important to note that when When that happens, the corresponding connection feature value is directly assigned the value 1.

[0058] When a driver encounters traffic congestion while driving, the first thing they will consider is to take a detour. In order to extract the actual spatial connection relationship between each detection point, an optional path matrix is ​​established based on the connection relationship between different detection points.

[0059] Specifically, the number of different paths between any two numbered detection points is counted. Based on all these counts, the number of different paths between the two numbered detection points is proportionally normalized. The result of the proportional normalization is used to construct an optional path matrix, where the numbered points are... The detection point arrived at the numbered The normalized result of the number of different paths to the detection point is the number of the first path in the optional path matrix. Line 1 The element values ​​of the column.

[0060] It is important to note that when When that happens, the number of different paths will be directly assigned to 0.

[0061] It is understandable that the larger the data in the optional path matrix, the more passage strategies can be selected between the two detection points corresponding to the values.

[0062] In order to capture the spatial correlation between different detection points, a traffic flow similarity matrix is ​​determined based on the correlation between the traffic flow vectors of different detection points at the same collection time.

[0063] In this context, the number of rows and columns in the traffic flow similarity matrix at the same collection time are both equal to the number of detection points. The traffic flow similarity matrix at the [missing information] time... Line 1 Column data For the number The detection points and their numbers are The correlation coefficient of the traffic flow vector at the same acquisition time for the detection points. , , This indicates the number of testing points.

[0064] Calculating the correlation coefficients of different vectors is a well-known technique and will not be elaborated further. In this embodiment, the Pearson correlation coefficient is selected as the correlation coefficient of the traffic flow vector. In practical applications, as other implementation methods, based on achieving the goal of calculating the correlation coefficients of different vectors, implementers may use other existing methods such as cosine similarity and Spearman correlation coefficient to calculate the correlation coefficients of different vectors. This application does not impose any special restrictions.

[0065] Feature fusion is performed on the connectivity feature matrix, optional path matrix, and traffic flow similarity matrix to extract traffic information containing road connectivity, spatial connectivity, and spatial correlation between detection points on the road to be predicted to be congested. Based on the connectivity feature matrix, optional path matrix, and traffic flow similarity matrix at each collection time, the fused traffic feature matrix for each collection time is determined.

[0066] The specific process is as follows:

[0067] The connection feature matrix, optional path matrix, and traffic flow similarity matrix at each data collection time are used as inputs. The Alternating Direction Multiplier Method (ADMM) is used to obtain the fusion result of these three matrices. This fusion result is then used as the fused traffic feature matrix at each data collection time. The fused traffic feature matrix at each acquisition time is represented as follows: The Alternating Direction Multiplier Method (ADMM) is a well-known technique, and its specific process will not be elaborated further.

[0068] It should be noted that the objective function of the Alternating Direction Multiplier Method (ADMM) in this application when obtaining the fused traffic feature matrix is... for:

[0069]

[0070] In the formula, Describes the minimum value function. This indicates the number of matrices used in the alternating direction multiplier method, in this application. The value is 3. It is the fusion result of the alternating direction multiplier method output. Indicates the first step of the alternating direction multiplier method. matrix, express and The difference metric distance is used to ensure that the elements of the fusion result have a certain degree of fidelity with the elements of the input matrix. The difference metric distance can include, but is not limited to, Euclidean distance, F-norm distance, etc. Indicates the first step of the alternating direction multiplier method. The weights of the matrices, The first for measuring input The influence of each matrix on the fusion result satisfies In this embodiment, the following settings are configured respectively. , , The empirical values ​​are 0.3, 0.3, and 0.4; The constraint term represents the constraint fusion of the traffic feature matrix. any element in The value is positive to ensure the uniqueness of the solution, and at the same time, to ensure that the weights are not negative when used as a weight matrix for subsequent prediction.

[0071] The flowchart for obtaining the fusion traffic feature matrix is ​​as follows: Figure 2 As shown.

[0072] At this point, the fused traffic feature matrix for each acquisition moment is obtained.

[0073] Step S003: Use a sine function to encode each collection time, obtain the time point encoding value of each collection time, and combine the traffic flow matrix and the fused traffic feature matrix at the collection time to obtain the traffic flow prediction data of each detection point at the next collection time after the collection time.

[0074] Traffic flow in urban road systems exhibits periodic variations with the natural day and week. For example, traffic flow is higher during the morning and evening rush hours on weekdays, and higher on weekdays than on non-weekdays. To more accurately extract this periodic trend, a sine function is used to encode the data collection time, and the time-point code value for each collection time is calculated using the following formula:

[0075]

[0076] In the formula, Indicates the first The time point encoding value of each acquisition moment; This indicates the number of data collection moments within one hour; This represents the sine function.

[0077] It is important to understand that the value of the time point code does not have a clear physical meaning. The purpose of the time point code is to describe the periodic characteristics of traffic flow through the periodic variation characteristics of the sine function.

[0078] Each traffic flow in the traffic flow matrix at the time of collection is updated to the traffic flow and the corresponding time point code value at the time of collection, with the traffic flow and time point code value separated by a comma, to obtain the updated traffic flow matrix at the time of collection.

[0079] It is understandable that updating the traffic flow matrix involves... Line 1 Column data For the first The detection point at the first Traffic flow at the first data collection time and the first data collection time The data is coded at each time point, with traffic flow separated from the coded time point by a comma. .

[0080] The updated traffic flow matrix at the time of collection is used as the data matrix, and the fused traffic feature matrix at the time of collection is used as the weight matrix. These are input into the STGCN deep learning framework to obtain the traffic flow prediction data of each detection point at the next collection time after the current collection time.

[0081] The STGCN deep learning framework uses the MSE loss function. Using the MSE loss function to obtain prediction data is a well-known technique and will not be elaborated further.

[0082] It is understandable that the next collection time after the collection time is the future collection time, and the traffic flow prediction data is the traffic flow prediction value at the future collection time.

[0083] At this point, traffic flow prediction data for each detection point at the next data collection time following the initial data collection time is obtained.

[0084] Step S004: Based on whether the data collection time falls within the weekday rush hour, traffic flow prediction data, and the width of the road where the detection point is located, traffic congestion prediction is achieved.

[0085] The time period weight of the collection time is assigned based on whether the collection time falls within the weekday commuting peak hours. Specifically, when the collection time falls within the weekday commuting peak hours, the time period weight of the collection time is assigned to the first preset value; when the collection time does not fall within the weekday commuting peak hours, the time period weight of the collection time is assigned to the second preset value.

[0086] Wherein, the first preset value and the second preset value are both preset parameter values. The first preset value should be greater than the second preset value. In this embodiment, the first preset value and the second preset value are 2 and 1, respectively. The commuting peak period includes the morning peak and the evening peak. The morning peak is from 7:00 to 9:00 and the evening peak is from 17:00 to 20:00.

[0087] The ratio of the time period weight of the data collection time to the width of the road where the detection point is located is recorded as the first ratio of the detection point at the data collection time. The positive correlation between the first ratio of the detection point at the data collection time and the traffic flow prediction data at the next data collection time after the data collection time is recorded as the traffic congestion index of the detection point at the data collection time.

[0088] It is understood that a positive correlation is applied to the first ratio at the time of data collection and the traffic flow prediction data at the next data collection time following the time of data collection at the detection point. This ensures that both the first ratio at the time of data collection and the traffic flow prediction data at the next data collection time are positively correlated with the traffic congestion index at the time of data collection at the detection point. It is understood that the positive correlation in this application refers to the relationship between the independent and dependent variables. The independent variables are the first ratio at the time of data collection and the traffic flow prediction data at the next data collection time following the time of data collection at the detection point, and the dependent variable is the traffic congestion index at the time of data collection at the detection point. A positive correlation means that the dependent variable increases (decreases) as the independent variable increases (decreases), and this can be an additive or multiplicative relationship.

[0089] Preferably, as an embodiment of this application, the product of the first ratio of the detection point at the collection time and the traffic flow prediction data at the next collection time after the detection point is recorded as the traffic congestion index of the detection point at the collection time.

[0090] The traffic congestion index at the collection time of the detection point is compared with the preset traffic congestion threshold: when the traffic congestion index is greater than the traffic congestion threshold, the detection point corresponding to the collection time of the traffic congestion index is marked as congested; when the traffic congestion index is less than or equal to the traffic congestion threshold, the detection point corresponding to the collection time of the traffic congestion index is marked as normal traffic.

[0091] This concludes the traffic congestion prediction.

[0092] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A method for predicting congestion in high-risk accident areas, characterized in that, The method includes the following steps: Extract the connectivity between different detection points, the width of the road, and the traffic flow at different collection times for different detection points. Number the detection points and determine the connection feature values ​​of different detection points based on the connectivity. For example, when two detection points are directly connected by a road and there are no other detection points between them, assign the connection feature value of the two detection points to 1. When two detection points are connected by a road and there are other detection points between them, the sum of the number of other detection points in the road connecting the two detection points and 1 is used as the connection feature value of the two detection points. Based on traffic flow, establish the traffic flow vector and traffic flow matrix for each detection point at each data collection time; Based on the identification numbers and connection feature values ​​of different detection points, a connection feature matrix is ​​established. Based on the connectivity relationships between different detection points, an optional path matrix is ​​established. The optional path matrix is ​​as follows: The number of rows and columns in the optional path matrix is ​​equal to the number of detection points. Line 1 Column data For the number The detection point arrived at the numbered The normalized value of the number of different paths at the detection point; , , Indicates the number of detection points; when When the time is right, the number of different paths will be set to 0. Based on the correlation between traffic flow vectors at different detection points at the same collection time, the traffic flow similarity matrix at the same collection time is determined. Based on the connection feature matrix, the optional path matrix, and the traffic flow similarity matrix at each collection time, the fused traffic feature matrix at each collection time is determined respectively. Each collection time is encoded using a sine function to obtain the time point encoding value of each collection time. Combined with the traffic flow matrix and the fused traffic feature matrix at each collection time, the traffic flow prediction data of each detection point at the next collection time after the collection time is obtained. Based on whether the data collection time falls within the weekday rush hour, traffic flow prediction data, and the width of the road where the monitoring point is located, traffic congestion prediction is achieved, including: The time period of the data collection time is weighted according to whether the data collection time falls within the peak commuting hours on a weekday. The ratio of the time period weight of the data collection time to the width of the road where the detection point is located is recorded as the first ratio of the detection point at the data collection time. The positive correlation between the first ratio of the detection point at the data collection time and the traffic flow prediction data at the next data collection time after the data collection time is recorded as the traffic congestion index of the detection point at the data collection time. When the traffic congestion index is greater than the preset traffic congestion threshold, the detection point corresponding to the collection time of the traffic congestion index is marked as congested; when the traffic congestion index is less than or equal to the preset traffic congestion threshold, the detection point corresponding to the collection time of the traffic congestion index is marked as normal traffic.

2. The congestion prediction method for high-risk accident areas according to claim 1, characterized in that, The specific method for establishing the traffic flow vector and traffic flow matrix for each detection point at each collection time based on traffic flow is as follows: Record any collection time as the target collection time. Arrange the traffic flow data collected at the same detection point at the target collection time and at each time interval before the target collection time into a column vector according to the time sequence, and obtain the traffic flow vector of the detection point at the target collection time. Arrange the traffic flow vectors of all detection points from left to right in ascending order of their numbers to obtain the traffic flow matrix of the detection points at the target collection time.

3. The congestion prediction method for high-risk accident areas according to claim 1, characterized in that, The connection feature matrix is ​​specifically: The number of rows and columns in the connection feature matrix is ​​equal to the number of detection points. The nth row in the connection feature matrix... Line 1 Column data For the number The detection points and their numbers are The connection feature values ​​of the detection points; when When the value is 1, the corresponding connection feature value is assigned.

4. The congestion prediction method for high-risk accident areas according to claim 1, characterized in that, The traffic flow similarity matrix at the same data collection time is specifically as follows: The number of rows and columns in the traffic flow similarity matrix at the same collection time is equal to the number of detection points. The number of rows and columns in the traffic flow similarity matrix is ​​the same as the number of detection points. Line 1 Column data For the number The detection points and their numbers are The correlation coefficient of the traffic flow vector at the same acquisition time for the detection points.

5. The congestion prediction method for high-risk accident areas according to claim 1, characterized in that, The formula for calculating the fused traffic feature matrix is: in, Indicates the first A fused traffic feature matrix for each data collection time; , and These represent the first preset parameter, the second preset parameter, and the third preset parameter, respectively, and the sum of the first preset parameter, the second preset parameter, and the third preset parameter is 1. Represents the connection feature matrix; Represents the optional path matrix; Indicates the first Traffic flow similarity matrix at each data collection time; This indicates the calculation of the Hadamard product.

6. The congestion prediction method for high-risk accident areas according to claim 1, characterized in that, The formula for calculating the time point encoding value at the acquisition time is: In the formula, Indicates the first The time point encoding value of each acquisition moment; This indicates the number of data collection moments within one hour; This represents the sine function.

7. The congestion prediction method for high-risk accident areas according to claim 1, characterized in that, The method for obtaining traffic flow prediction data for each detection point at the next collection time after the collection time by combining the traffic flow matrix at the collection time and the fused traffic feature matrix includes the following specific methods: Update each traffic flow in the traffic flow matrix at the time of collection to the traffic flow and the corresponding time point code value at the time of collection, with the traffic flow and the time point code value separated by a comma, and obtain the updated traffic flow matrix at the time of collection. Using the updated traffic flow matrix at the time of collection as the data matrix and the fused traffic feature matrix at the time of collection as the weight matrix, the traffic flow prediction data of each detection point at the next collection time after the current collection time is obtained according to the STGCN deep learning framework.

Images