Method for identifying dangerous road section of long-distance passenger traffic based on vehicle-mounted safety alarm data

Abstract

The invention belongs to the technical field of traffic safety, and particularly relates to a long-distance passenger traffic dangerous road section identification method based on vehicle-mounted safety alarm data. The method specifically comprises the following steps: inputting longitude and latitude data point sequences of a long-distance passenger transport driving alarm position, and automatically adjusting DBSCAN clustering algorithm parameters according to data characteristics of the data point sequences to replace assignment parameters based on artificial experience. A DBSCAN clustering algorithm is used for processing the data point sequence to obtain a set of a plurality of groups of alarm position data points which are sufficiently and closely gathered together, and the set corresponds to a dangerous road section range needing to be identified in a geographic space; the automatically-adjusted DBSCAN clustering algorithm parameters comprise the steps of searching the radius r of adjacent data points and judging the minimum number nm of data points which are dense enough within the radius range. According to the method, parameters are automatically adjusted through index calculation and an iterative algorithm, experience driving is replaced by data driving, universal and accurate dangerous road section recognition is achieved, and therefore the road traffic safety condition is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of traffic safety, and in particular relates to a method for identifying dangerous road sections for long-distance passenger vehicles based on vehicle-mounted safety alarm data. Background Art

[0002] Ensuring driving safety is a major concern in the field of road traffic. To this end, on-board safety alarm systems have been widely used, which can issue safety alarms in real time based on abnormal vehicle operating conditions and driver conditions. However, isolated on-board safety alarm systems often have problems such as invalid repeated reminders, false alarms, and missed alarms, which make experienced drivers pay less attention to their alarms, and novice drivers cannot effectively use alarm records to improve driving safety on dangerous roads. This has led to the current on-board alarm system being ineffective and unable to substantially improve driving safety. In particular, long-distance passenger vehicles are generally equipped with on-board alarm systems, and the routes are relatively fixed. If the alarm data can be used to identify typical dangerous sections of the route, it will be of great significance to improve driving safety; however, current research and literature have not yet touched on this.

[0003] Existing methods for identifying dangerous sections of roads mostly use traffic accident or vehicle motion data and rely on empirical thresholds, and the objectivity of the identification results needs to be improved. Existing methods for identifying dangerous sections of roads based on traffic accident or vehicle motion data often require the support of additional data, parameters, and manual experience, which is inconvenient to use. For example, the patent document with application publication number CN 115691182 A proposes "a method, device, and computer storage medium for identifying potentially dangerous sections of roads", which determines the degree of danger of a section of road through the location of abnormal driving events (such as sharp turns, abnormal bumps, tire slippage, locking, etc.), vehicle credit scores, and driving risk coefficients, but the vehicle credit scores need to be additionally confirmed. Another example is the patent document with the authorization publication number CN109493600 B, which proposes a "method for identifying sections with high traffic accident rates based on accident risk". The risk is calculated by the number of casualties and road speed limits of historical accidents under certain time, space and environmental conditions, and the locations with risk greater than the threshold are identified as accident-prone points. Similar to the traditional identification method of counting the number of alarms in a section, the identification threshold used has a significant impact on the result. However, the determination of the threshold lacks a theoretical basis, which can easily lead to the identification result being inconsistent with the actual situation. There are also academic papers that propose identification methods based on statistical knowledge such as spatial autocorrelation analysis, but the studied section must first be divided into small sections according to certain standards. The division standard is also determined by experience. Improper division will significantly affect the accuracy and precision of the results, such as truncating a continuous dangerous section into two, and including both general sections and dangerous sections in an overly long "dangerous section"... These may affect the accurate identification of dangerous sections and the analysis and research of their causes. Therefore, it is urgent to propose a more universal method to reduce the dependence on manual experience, and it can be widely used in data sets of different spatial forms. Summary of the invention

[0004] In view of the fact that the vehicle alarm system has not yet fully exerted its potential, and the current dangerous road section identification method has the limitations of input data, the dependence on manual experience and the ambiguity of spatial identification, the present invention aims to propose a method that can automatically identify the dangerous road sections of long-distance passenger vehicles using the safety alarm data set, and automatically adjust the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm parameters to avoid the influence of experience-based parameters on the recognition effect; by clustering the longitude and latitude of the alarm location, the accurate range of the dangerous road section is identified to avoid the influence of mechanical road section division on the recognition accuracy. This method is universal and accurate, and can help improve traffic safety.

[0005] To achieve the above objectives, the present invention proposes a DBSCAN clustering algorithm that effectively utilizes existing vehicle-mounted alarm data and can automatically adjust algorithm parameters, thereby identifying dangerous sections for long-distance passenger vehicles.

[0006] Technical solution of the present invention:

[0007] A method for identifying dangerous sections of long-distance passenger vehicles based on vehicle-mounted safety alarm data, which inputs a sequence of longitude and latitude data points of the alarm position of long-distance passenger vehicles, and automatically adjusts the parameters of the DBSCAN clustering algorithm according to the data characteristics of the data point sequence, instead of assigning parameters based on manual experience. The DBSCAN clustering algorithm processes the data point sequence to obtain a set of several groups of alarm position data points that are closely clustered together (called a cluster), which corresponds to the range of dangerous sections to be identified in the geographic space.

[0008] The parameters of the DBSCAN clustering algorithm that are automatically adjusted include: the radius r of searching for neighboring data points and the minimum number of data points n within the radius to determine whether the data points are dense enough m .

[0009] During the operation of the general DBSCAN clustering algorithm, these two parameters are assigned by manual experience, which lacks scientific basis and cannot guarantee the recognition effect. The same set of parameters is also difficult to use universally. The present invention will automatically adjust these two parameters through indicator calculation and iterative algorithm, replacing experience-driven with data-driven, to achieve universal and accurate identification of dangerous road sections.

[0010] Furthermore, a method for identifying dangerous sections of long-distance passenger vehicles based on vehicle safety alarm data specifically includes the following steps:

[0011] Step 1: Initialize the DBSCAN clustering algorithm parameters, set the search radius r = 1 meter, and the minimum number of data points n m = 2. Set a smaller initial value to leave enough room for parameter adjustment.

[0012] Step 2: Use the DBSCAN clustering algorithm to perform clustering and remove isolated points that have not been successfully clustered to obtain several groups of valid clusters consisting of multiple data points.

[0013] Step 3: Automatically find clustering algorithm parameters through indicator calculation and iterative algorithm, without relying on blind experience assignment, to ensure that the clustering algorithm can finally obtain several clusters with significantly denser data points. The specific process is as follows:

[0014] 3.1: Calculate the average distance from each point in each valid cluster to the cluster center, denoted as d.

[0015] 3.2: Divide the number of data points n of each cluster by the average distance d to obtain the pseudo-density index p.

[0016] 3.3: Arrange each cluster in descending order according to the index p. Denote the cluster with the largest p value as the first cluster, the cluster with the second largest p value as the second cluster, and so on, and mark the p value ordinal numbers of all valid clusters. Subsequently, through differential processing, find the cluster with the largest difference in p value from the subsequent cluster, and denote its ordinal number as i. At this time, the p values of the first i clusters will be significantly greater than those of other clusters, that is, the distribution of alarm data in the first i clusters is denser, and the alarm events in the corresponding spatial range are more concentrated. Therefore, it is speculated that the corresponding road sections are more dangerous.

[0017] 3.4: Under the condition of limited resources, consider preferentially disposing of the most prominent k dangerous road sections. Therefore, compare i and k, record the current i value closest to k and the corresponding clustering parameters, and denote this set {i, r, n m} as S. The parameters in the set S can be used to reproduce the existing optimal clustering results, which is a compressed way of saving the optimal results and can ensure that the final results will not deteriorate at a relatively low storage cost.

[0018] ① If i = k, terminate the algorithm, output S and the k clustering clusters with the largest p values, map the data points in these clustering clusters to the road space, and the obtained k road space ranges are the most prominent k dangerous road sections;

[0019] ② If i < k and S is updated, continue to adjust the algorithm parameters adjusted last time;

[0020] ③ If i < k and S is not updated for the first time, roll back one clustering parameter and adjust another algorithm parameter, that is: if the previous parameter adjustment is r, let n m = n m + 1 to increase the cluster density and reduce the average distance d from the data points in the cluster to the cluster center; if the previous parameter adjustment is n m or just completed the initial clustering, take out the element i from the current set S, denote it as i s , and let

[0021] increase the number of data points n in the cluster;

[0022] ④ If i < k and S is not updated for the second time or more, randomly adjust the parameters r or n m , and when S is not updated continuously for L (an upper limit number determined by oneself, the fewer the number of times, the faster the algorithm converges; the more the number of times, the more sufficient the algorithm searches), terminate the algorithm, output the current S and the i s clustering clusters with the largest index p, map the data points in these clusters to the road space, and the obtained i s road space ranges are the most prominent i s dangerous road sections;

[0023] ⑤If i>k, then let Expand the search range, increase the number of data points n in each cluster, and reduce the total number of clusters; when S has not been updated for L consecutive times, terminate the algorithm and output the current S and the i with the largest index p s clusters, and correspond the data points in these clusters to the road space. s The spatial range of the road segment is the most prominent i s Dangerous road section.

[0024] 3.5: According to common sense, the search radius r cannot be increased indefinitely. Two dangerous sections that are too far apart (such as beyond the driving field of vision) should be treated independently, and an upper limit value r is set for r. m , when r reaches r m After that, only n is allowed to increase m .

[0025] Step 4: Repeat steps 2 to 3.5 until the algorithm is terminated.

[0026] Beneficial Effects

[0027] After applying the above technical solutions, the present invention can achieve the following beneficial effects: in the general DBSCAN clustering algorithm operation process, the DBSCAN algorithm parameters are assigned by manual experience, lacking scientific basis, unable to guarantee the recognition effect, and the same set of parameters is difficult to be widely used; the present invention automatically adjusts the parameters through index calculation and iterative algorithm, replaces experience-driven with data-driven, and realizes universal and accurate identification of dangerous sections. The DBSCAN clustering algorithm parameters can be automatically adjusted based on the characteristics of the safety alarm data set used in view of the randomness of the spatial morphology of dangerous sections, avoiding errors caused by manual experience-assigned parameters, and identifying k prominent dangerous sections of any shape. The universality and accuracy of the method are improved compared with traditional methods, and can improve road traffic safety in terms of safety alarm systems and scientific research. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 Automatically adjust the DBSCAN algorithm parameters r and n for the present invention m , the algorithm flow chart for identifying k prominent dangerous sections;

[0029] Figure 2 This is the basic process of the original DBSCAN algorithm. The present invention improves the process of "setting algorithm parameters";

[0030] Figure 3 The first 8 rows of a p-value in descending order obtained by the embodiment based on the own data set are easy to find that i=1 at this time;

[0031] Figure 4 This is a schematic diagram of a dangerous road section output by the embodiment. DETAILED DESCRIPTION

[0032] The technical solution provided by the present application will be further described below in conjunction with specific embodiments and accompanying drawings. The advantages and features of the present application will become more apparent with the following description.

[0033] A method for identifying dangerous sections of long-distance passenger buses based on on-board safety alarm data, which inputs a sequence of longitude and latitude data points of the alarm position of the long-distance passenger bus, and automatically adjusts the DBSCAN clustering algorithm parameters according to the data characteristics. Using the parameters, the clustering algorithm obtains a set of several groups of alarm position data points that are clustered closely together, which corresponds to the range of dangerous sections to be identified.

[0034] The basic process of the original DBSCAN algorithm is as follows Figure 2 As shown, the DBSCAN algorithm can cluster data point sequences without being constrained by spatial morphology, but is sensitive to algorithm parameters. General DBSCAN algorithm parameters are assigned by manual experience, lack scientific basis, are difficult to use, and are difficult to ensure the algorithm operation effect. According to the input alarm location data set, the present invention can automatically adjust the algorithm parameters to obtain densely distributed clusters of alarm data in any form, and the corresponding geographic space range is the dangerous section.

[0035] The parameters of the DBSCAN clustering algorithm that are automatically adjusted include: the radius r of searching for neighboring data points and the minimum number of data points n within the radius to determine whether the data points are dense enough m ;

[0036] The identification process specifically includes the following steps: Figure 1 (shown)

[0037] Step 1: Initialize the DBSCAN clustering algorithm parameters, set the search radius r = 1 meter, and the minimum number of data points n m =2.

[0038] Step 2: Use the DBSCAN clustering algorithm to perform clustering and remove isolated points that have not been successfully clustered to obtain several groups of valid clusters consisting of multiple data points.

[0039] Step 3: Automatically find clustering algorithm parameters through indicator calculation and iterative algorithm, without relying on blind experience assignment, to ensure that the clustering algorithm can finally obtain several clusters with significantly denser data points.

[0040] 3.1: Calculate the average distance from each point in each valid cluster to the cluster center, denoted as d.

[0041] 3.2: Divide the number of data points \(n\) of each type of cluster by the average distance \(d\) to obtain the pseudo-density index \(p\).

[0042] 3.3: According to the index \(p\), sort each cluster in descending order, and find the position with the largest difference in the \(p\) value from the next cluster through differential processing, and record its ordinal number as \(i\).

[0043] As Figure 3 shown: The first 8 rows of a sequence sorted in descending order of \(p\) values obtained from the self-owned dataset. Before the comma in each row is the cluster index obtained by the clustering algorithm, and after the comma is the \(p\) value of this cluster; It can be seen from the figure that the difference between the \(p\) value of the first cluster, 42.1339, and the \(p\) value of the second cluster, 21.0669, is the largest. Therefore, at this time \(i = 1\).

[0044] 3.4: In the case of limited resources, select the top \(k\) most prominent dangerous road sections for priority handling (here \(k = 10\) is selected), compare \(i\) and \(k\), record the current \(i\) value closest to \(k\) and the corresponding clustering parameters, and denote this set \(\{i, r, n\) m \} as \(S\).

[0045] ① If \(i = k\), terminate the algorithm, output \(S\) and the \(k\) clusters with the largest \(p\) values, and map the data points in these clusters to the road space. The obtained \(k\) road space ranges are the top \(k\) most prominent dangerous road sections;

[0046] ② If \(i < k\) and \(S\) is updated, continue to adjust the algorithm parameters adjusted last time;

[0047] ③ If \(i < k\) and \(S\) is not updated for the first time, roll back one clustering parameter and adjust another algorithm parameter, that is: if the previous adjusted parameter is \(r\), let \(n\) m = n m + 1 to increase the cluster density and reduce the average distance \(d\) from the data points in the cluster to the cluster center; if the previous adjusted parameter is \(n\) m or just completed the initial clustering, take out the element \(i\) from the current set \(S\), denoted as \(i\) s , \(i\) s is the serial number of the cluster with the largest difference from the next cluster after sorting all clusters in descending order according to the pseudo-density index \(p\) value. Let increase the number of data points \(n\) in the cluster;

[0048] ④ If \(i < k\) and \(S\) is not updated for the second time and above, randomly adjust the parameter \(r\) or \(n\) m , when \(S\) is not updated continuously for \(L\) (the upper limit number determined by oneself, here \(L = 50\) is selected) times, terminate the algorithm, output the current \(S\) and the largest \(i\) s clusters of the index \(p\), and map the data points in these clusters to the road space. The obtained \(i\) s road space ranges are the top \(i\) most prominents On dangerous roads;

[0049] ⑤If i>k, then let Increase the number of data points n in the cluster and reduce the total number of clusters; when S has not been updated for L consecutive times, terminate the algorithm and output the current S and the i with the largest index p s clusters, and correspond the data points in these clusters to the road space. s The spatial range of the road segment is the most prominent i s Dangerous road section.

[0050] 3.5: According to common sense, the search radius r cannot be increased indefinitely, and a certain upper limit value r is set for it m (According to the general vision during driving, select r here m = 200 m), when r reaches r m After that, only n is allowed to increase m .

[0051] Step 4: Repeat steps 2 to 3.5 until the algorithm is terminated.

[0052] Part of the process of cyclic update of algorithm parameters is shown in Table 1:

[0053] Table 1 Part of the process of cyclic update of algorithm parameters in this embodiment

[0054]

[0055]

[0056] In the own data set, because the set S has not been updated for 50 consecutive times, the algorithm terminates. At this time, r = 200 meters, n m =11, 9 prominent dangerous sections were output.

[0057] like Figure 4 The figure shows a dangerous road section sign in the final output.

[0058] The above description is only a description of the preferred embodiments of the present application, and is not intended to limit the scope of the present application. Any changes or modifications made by any person skilled in the art based on the above disclosed technical contents shall be deemed as equivalent effective embodiments and shall fall within the scope of protection of the technical solution of the present application.

Claims

1. A method for identifying dangerous sections of long-distance coach driving based on on-vehicle safety alarm data. Input the longitude and latitude data point sequence of the alarm locations of long-distance coach driving, and automatically adjust the parameters of the DBSCAN clustering algorithm according to the data characteristics of the data point sequence to replace the parameter assignment based on manual experience. The DBSCAN clustering algorithm processes the data point sequence to obtain several sets of alarm location data points that are closely clustered together, which is called a cluster, and corresponds to the range of the dangerous section to be identified in the geographical space; The automatically adjusted DBSCAN clustering algorithm parameters include: Search the radius r of the neighboring data points and determine the minimum number of data points n within the radius that are dense enough m .

2. A method for identifying dangerous sections of long-distance passenger vehicles based on vehicle safety alarm data as claimed in claim 1, characterized in that: The identification method includes the following steps: Step 1: Initialize the parameters of the DBSCAN clustering algorithm; Step 2: Use the DBSCAN clustering algorithm for clustering, and eliminate the isolated points that are not successfully clustered to obtain several effective clusters composed of multiple data points; Step 3: Automatically find the parameters of the clustering algorithm through index calculation and iterative algorithm, without relying on blind empirical assignment, to ensure that several clusters with significantly denser data points can be obtained through the clustering algorithm finally; the specific process is as follows: 3.1: Calculate the average distance from each point in each effective cluster to the cluster center, denoted as d; 3.2: Divide the number of data points n in each cluster by the average distance d to obtain the pseudo-density index p; 3.3: According to the index p, sort all the clusters in descending order. Denote the cluster with the largest p value as the first cluster, the cluster with the second largest p value as the second cluster, and so on, and mark the p value ordinal numbers of all effective clusters. Subsequently, find the cluster with the largest difference in p value from the subsequent cluster through differential processing, and denote its ordinal number as i. At this time, the p values of the first i clusters will be significantly greater than those of other clusters, that is, the distribution of alarm data in the first i clusters is denser, and the alarm events in the corresponding spatial range are more concentrated, so it is speculated that the corresponding section is more dangerous; 3.4: In the case of limited resources, we consider giving priority to the most prominent k dangerous sections, so we compare i and k, record the current i value closest to k and the corresponding clustering parameters, and convert this set {i, r, n m } is denoted as S. The parameters in set S can be used to reproduce the existing optimal clustering results. It is a compressed way to save the optimal results, which can ensure that the final results will not be degraded at a low storage cost; ① If i = k, terminate the algorithm, output S and the k clusters with the largest p values, and map the data points in these clusters to the road space. The k road space ranges obtained are the most prominent k dangerous sections; ② If i < k and S is updated, continue to adjust the algorithm parameters adjusted last time; ③ If i < k and S is not updated for the first time, roll back the clustering parameters once and adjust another algorithm parameter, that is: if the previous parameter adjustment is r, let n m = n m + 1 to increase the cluster density and reduce the average distance d from the data points within the cluster to the cluster center; if the previous parameter adjustment is n m or just after the initial clustering is completed, take out the element i from the current set S, denoted as i s , and let increase the number of data points n within the cluster; ④ If i < k and S has not been updated for the second time or more, randomly adjust the parameters r or n m , when S has not been updated continuously for L times, terminate the algorithm and output the current S and the i with the largest index p s clusters, map the data points in these clusters to the road space, and the obtained i s section of the road space range is the most prominent i s is the dangerous road section; L is the upper limit number set by oneself. The fewer the number of times, the faster the algorithm converges; the more the number of times, the more sufficient the algorithm searches; ⑤If i>k, then let Expand the search range, increase the number of data points n in each cluster, and reduce the total number of clusters; when S has not been updated for L consecutive times, terminate the algorithm and output the current S and the i with the largest index p s clusters, and correspond the data points in these clusters to the road space. s The spatial range of the road segment is the most prominent i s On dangerous roads; 3.5: According to common sense, the search radius r cannot be increased indefinitely. Two dangerous sections that are too far apart should be treated independently. An upper limit value r is set for r. m , when r reaches r m After that, only n is allowed to increase m . Step 4: Repeat steps 2 to 3.5 until the algorithm is terminated.

3. A method for identifying dangerous sections of long-distance passenger vehicles based on vehicle safety alarm data as claimed in claim 2, characterized in that: In step 1, the search radius r = 1 meter and the minimum number of data points n m =2.

Images