TWI786902B - Device, method and computer program product for exploration of potential event hotspot - Google Patents

Abstract

A device, a method, a computer program product for exploration of potential event hotspot are provided. A plurality of exploration target objects, respective characteristics of the plurality of exploration target objects, and respective events of the plurality of exploration target objects are collected. The plurality of exploration target objects are grouped according to the respective characteristics of the plurality of exploration target objects. Exploration target objects in each group are divided into areas according to the respective events of the plurality of the exploration target objects. Exploration target objects in each area in the each group each group in each group are explored according to a risk index and time of occurrence, so as to obtain representative event characteristics of the exploration target objects in the each area in the each group. The representative event characteristics of the exploration target objects in the each area in the each group are compared with each other such that an exploration target object having potential event hotspot is obtained.

Description

Apparatus, method and computer program product for potential event hot spot detection

The present invention relates to a data processing and exploration technology, and in particular, relates to equipment for potential event hotspot exploration, a method for its execution, and a computer program product.

With the advancement of technology, the use of data to analyze the trend of road accidents has increased. The conventional method is to capture the time, weather, location, etc. of the road where the accident has occurred to analyze the degree of danger, thereby reminding the driver of those accidents. Road sections are accident hotspots. However, early warning cannot be issued for road sections where accidents have not occurred at all.

Therefore, how to conduct data mining to quickly and accurately find potential events is a problem to be solved in the industry.

In order to solve the above problems and other problems, the present invention discloses a potential event hotspot detection device, its execution method and computer program product.

The equipment for potential event hotspot exploration of the present invention includes: an exploration target data collection module, which collects multiple exploration target objects, the respective characteristics of the multiple exploration target objects, and the respective events of the multiple exploration target objects; According to the respective characteristics of the plurality of prospecting objects, the plurality of prospecting objects are grouped; the event feature mining module partitions the prospecting objects in each group according to the respective events of the plurality of prospecting objects, According to the risk indicators and time of occurrence, survey the exploration target objects in each area in each group, and then obtain the representative event characteristics of the exploration target objects in each area in each group; and The event feature comparison module compares the representative event features of the prospecting targets in each region in each group to obtain prospecting targets with potential event hotspots.

In one embodiment, the grouping module selects a plurality of prospecting objects from the plurality of prospecting objects as center points, and calculates the relationship between the unselected prospecting objects and the plurality of centers according to the characteristics of the plurality of prospecting objects. The respective distances between the points, so as to divide the unselected prospecting target object and the nearest central point into the same group, and then calculate the distance between each prospecting target object in the same group and other prospecting target objects in the same group The distance sum of , select the prospecting target object with the minimum distance sum as the new center point in the same group, and then recalculate with the new center point until the new center point does not change.

In one embodiment, the event feature mining module partitions the mining target objects in each group according to whether the respective events of the plurality of mining target objects have occurred the target event, so as to obtain the number of events before the target event occurs A time interval is used as a window to sample the respective events of the survey target objects that have occurred the target event, and the respective events of the survey target objects that have not occurred the target event are sampled using the time interval as a sliding window.

In one embodiment, the event feature detection module uses the risk index with a length of 1 and its corresponding time interval to detect the sampled events of the detection target object of the target event that has occurred, and obtain the target event that has occurred The effective event characteristics of the exploration target object, and then gradually increase the length of the risk index, to obtain the representative event characteristics of the exploration target object that has occurred the target event, and the length of the risk index is 1 and its corresponding time interval Prospecting for the sampled events of the exploration target object that has not occurred the target event, obtaining the respective effective event characteristics of the exploration target object that has not occurred the target event, and then gradually increasing the length of the risk index to obtain the non-occurrence target event The respective representative event characteristics of the prospecting target objects.

In one embodiment, the event feature comparison module evaluates the risk of the target event occurring on the prospecting target object with potential event hotspots for the prospecting target object with potential event hotspots.

In one embodiment, the plurality of prospecting objects are plural road data, the respective characteristics of the plurality of prospecting objects are the respective road characteristics of the plurality of road data, and the respective events of the plurality of prospecting objects include A set consisting of vehicle number, hazard indicator, return time, and whether an accident occurred.

In one embodiment, the device of the present invention further includes: an event hotspot early warning module, which issues an early warning to the vehicle according to the base station identification code corresponding to the road data of the hot spots of the accidents and potential accidents.

The method for potential event hotspot detection performed by the device of the present invention includes: collecting a plurality of exploration target objects, respective characteristics of the plurality of exploration target objects, and respective events of the plurality of exploration target objects; according to the plurality of exploration target objects According to the respective characteristics of the target objects, the plurality of prospecting target objects are grouped; according to the respective events of the plurality of prospecting target objects, the According to the risk index and occurrence time, the exploration target objects in each area in each group are explored, and the representative data of the exploration target objects in each area in each group are obtained. and comparing the representative event characteristics of the exploration target objects in each area in each group to obtain exploration target objects with potential event hotspots.

In one embodiment, the grouping includes: (1) selecting a plurality of prospecting objects from the plurality of prospecting objects as center points; (2) calculating the unselected The respective distances between the exploration target object and the plurality of center points, so that the unselected exploration target object and the nearest center point are divided into the same group; (3) calculate the distance between each exploration target object in the same group and The sum of distances between other prospecting target objects in the same group, to select the prospecting target object with the minimum distance sum as the new central point in the same group; and (4) repeat steps (2) and (3), until the new center point no longer changes.

In one embodiment, the partitioning includes: partitioning the prospecting target objects in each group according to whether the respective events of the plurality of prospecting target objects have occurred a target event; for the prospecting target objects that have occurred a target event, Taking a predetermined time interval before the occurrence of the target event as a window, sampling the respective events of the prospecting target objects that have occurred the target event; and using the predetermined time interval as a sliding window for the prospecting target objects that have not occurred the target event , to sample the respective events of the prospecting target object for which no target event occurs.

In one embodiment, the prospecting includes: for the sampled events of the prospecting target object of the target event that has occurred, perform prospecting with a risk indicator with a length of 1 and its corresponding time interval, and obtain the target event that has occurred Exploring the effective event characteristics of the target object, and then gradually increasing the length of the risk index, to obtain the representative event characteristics of the exploration target object that has occurred the target event; event, with length 1 The risk indicators and their corresponding time intervals are explored to obtain the respective effective event characteristics of the exploration target objects that have not occurred the target event, and then gradually increase the length of the risk index to obtain the respective effective event characteristics of the exploration target objects that have not occurred the target event. representative event features.

In one embodiment, the comparison includes: with respect to the prospecting target object with potential event hotspots, evaluating the risk of occurrence of a target event for the prospecting target object with potential event hotspots.

In one embodiment, the plurality of prospecting objects are plural road data, the respective characteristics of the plurality of prospecting objects are the respective road characteristics of the plurality of road data, and the respective events of the plurality of prospecting objects include A set consisting of vehicle number, hazard indicator, return time, and whether an accident occurred.

In one embodiment, the method includes: issuing an early warning to the vehicle according to the base station identification code corresponding to the road data corresponding to the hotspots of the accidents and potential accidents.

The computer program product of the present invention executes the above-mentioned method for potential event hotspot detection after being loaded into the program by the computer.

In other words, the equipment for detecting potential event hotspots and its execution method, as well as the computer program product of the present invention can detect the event characteristics based on the event occurrence frequency, occurrence time relationship, occurrence time interval and range of the target objects of the detection, and then According to the representative event characteristics of the exploration target objects that have occurred target events, find out the exploration target objects that have not occurred target events but have potential event hotspots, and give warnings to the exploration target objects that have occurred target events and have potential event hot spots , so as to effectively reduce the frequency of target events.

2: Equipment for hotspot detection of potential events

21: Exploration target data collection module

22: Grouping module

23:Event Feature Exploration Module

24: Event feature comparison module

25:Event hotspot early warning module

S201~S205: steps

S301~S306: steps

S401~S408: steps

FIG. 1 is a schematic diagram of the structure of the equipment for potential event hotspot detection according to the present invention.

FIG. 2 is a schematic flow chart of the method for potential event hotspot detection according to the present invention.

FIG. 3 is a schematic flow chart of the grouping of the method for potential event hotspot detection of the present invention.

FIG. 4 is a schematic flow chart of event feature detection and comparison of the method for potential event hotspot detection in the present invention.

The implementation of this case is described below through specific examples, and those skilled in this art can easily understand other advantages and effects of this case from the content disclosed herein. The structures, proportions, sizes, etc. shown in the drawings attached to this manual are only used to match the content disclosed in the manual, for the understanding and reading of those who are familiar with this technology, and are not used to limit the conditions that can be implemented in this case. Therefore, any modifications, changes or adjustments should still fall within the scope covered by the technical content disclosed in this case without affecting the functions and goals that can be achieved in this case.

Please refer to Fig. 1, the equipment (such as server) 2 of the potential event hotspot exploration of the present invention comprises exploration target data collection module 21, grouping module 22, event feature exploration module 23, event feature comparison module 24, event Hot spot warning module 25.

In one embodiment, each module in FIG. 1 can be software, hardware or firmware; if it is hardware, it can be a processing unit, processor, computer or server with data processing and computing capabilities; In the case of software or firmware, it may include instructions executable by a processing unit, processor, computer or server.

The prospecting target data collection module 21 is used to collect multiple prospecting target objects, respective features of the plurality of prospecting target objects, and respective events of the plurality of prospecting target objects, so as to collect the plurality of prospecting target objects, the respective features of the plurality of prospecting target objects, Features, each of the multiple prospecting target objects The events of are input to a database (not shown). In addition, the collection method is, for example, manual input or interfaced with an external system for automatic inclusion.

The grouping module 22 is used for grouping the plurality of prospecting objects according to their respective characteristics. In detail, the grouping module 22 can select a plurality of prospecting objects from the plurality of prospecting objects as center points, and calculate the distance between the unselected prospecting objects and the plurality of center points according to the characteristics of the plurality of prospecting objects. to divide the unselected prospecting target object into the same group with the closer or nearest central point, and then calculate the distance between each prospecting target object in the same group and other prospecting target objects in the same group The sum of the distances between them, and then select the prospecting target object with the smallest sum of distances as the new center point in the same group, and then recalculate with the new center point until the obtained center point does not change, that is, the grouping is completed.

The event feature prospecting module 23 partitions the prospecting target objects in each group according to the respective events of the plurality of prospecting target objects, so that the prospecting of each area in each group is The target object is surveyed, and then the representative event characteristics of the survey target object in each area in each group are obtained. Specifically, the event feature exploration module 23 can partition the exploration target objects in each group according to whether the respective events of the plurality of exploration target objects have occurred the target event, so that a time before the target event occurs The interval is a window to sample the respective events of the prospecting target objects that have occurred the target event, and use the time interval as a sliding window to sample the respective events of the prospecting target objects that have not occurred the target event. In addition, the event feature exploration module 23 further uses the risk index with a length of 1 and its corresponding time interval to explore the sampled events of the exploration target object of the target event that has occurred, and obtains the detection target object of the target event that has occurred Then gradually increase the length of the hazard index to obtain the representative event characteristics of the exploration target object that has occurred the target event, and the hazard index with a length of 1 and its corresponding time Prospecting for the sampled events of the exploration target object that has not occurred the target event, and obtaining the respective effective event characteristics of the exploration target object that has not occurred the target event, and then gradually increasing the length of the risk index, to obtain the non-occurrence Prospecting for Target Events The respective representative event characteristics of the target objects.

The event feature comparison module 24 is used for comparing the representative event features of the survey target objects in each region in each group, so as to obtain the survey target objects with potential event hotspots. In addition, the event feature comparison module 24 further evaluates the risk of the target event occurring on the prospecting target object with potential event hotspots for the prospecting target object with potential event hotspots.

The event hotspot early warning module 25 is used to issue an early warning to the vehicle according to the base station identification code corresponding to the road data corresponding to the hot spots of accidents and potential accidents. Specifically, the plurality of prospecting objects are plural road data, the respective characteristics of the plurality of prospecting objects are the respective road characteristics of the plurality of road data, and the respective events of the plurality of prospecting objects include vehicles A set consisting of machine number, hazard indicator, return time, and whether an accident occurred. Therefore, the event hotspot early warning module 25 can use the base station identification code to issue early warnings to vehicles near roads where accidents have occurred (have occurred) or roads where potential accident hotspots have occurred.

Please refer to Fig. 2, the method for potential event hotspot exploration of the present invention is a computer or server or information processing device or similar device equipment, through the coordinated operation of computer software and hardware resources, in which the program or instruction is operated to The execution method includes steps S201-S205. In addition, a computer program product is provided, and the method for potential event hotspot detection is executed after the program is loaded into the computer. It should be noted that computer software can be stored in computer-readable recording media such as hard disks, floppy disks, optical disks, and USB flash drives, and can also be directly transmitted and provided on the Internet, such as computer programs or computer program products.

In step S201, collect the plurality of mining target objects, the respective features of the plurality of mining target objects, and the respective events of the plurality of mining target objects, and then proceed to step S202.

In step S202, group the plurality of prospecting objects according to their respective characteristics, and then proceed to step S203.

In step S203, according to the respective events of the plurality of prospecting target objects, the prospecting target objects in each group are partitioned, and then step S204 is entered.

In step S204, according to the risk index and the time of occurrence, the exploration target object in each area in each group is surveyed, and the representative event characteristics of the exploration target object in each area in each group are obtained, and then Go to step S205.

In step S205, the representative event features of the survey targets in each area in each group are compared with each other to obtain survey targets with potential event hotspots.

Please refer to FIG. 3 , the grouping method in the method for detecting potential event hotspots of the present invention includes steps S301-S306.

In step S301, a plurality of prospecting target objects are selected as center points from the plurality of prospecting target objects, and then proceed to step S302.

In step S302, according to the respective characteristics of the plurality of prospecting objects, the respective distances between the unselected prospecting objects and a plurality of center points are calculated, so as to associate the unselected prospecting objects with the closer or closest distances The central points are divided into the same group, and then proceed to step S303.

In step S303, calculate the distance sum between each prospecting target object in the same group and other prospecting target objects in the same group, so as to select the prospecting target object with the minimum distance sum as the new center point in the same group, and then carry out Go to step S304.

In step S304, it is judged whether the new center point is the same as the old center point, if they are the same, then proceed to step S306 to complete the grouping, otherwise, if not, then proceed to step S305.

In step S305, replace the old center point with the new center point, and then return to step S302 until the center point does not change any more.

Please refer to FIG. 4 , the event feature detection and comparison method in the method for potential event hotspot detection of the present invention includes steps S401-S408.

In step S401, according to whether the respective events of the plurality of prospecting target objects have occurred the target event, the prospecting target objects in each group are partitioned, wherein, one zone is the prospecting target object that has (once) occurred the target event, Proceed to steps S402-S404, and the other area is the prospecting target object where no target event occurs, then proceed to steps S405-S407.

In step S402, for the prospecting target objects where the target event has occurred, a predetermined time interval before the target event occurs is used as a window to sample the respective events of the prospecting target objects where the target event has occurred, and then proceed to step S403.

In step S403, for the sampled events of the prospecting target objects in which the target event has occurred, the risk index with a length of 1 and its corresponding time interval are used for prospecting to obtain the effective event characteristics of the prospecting target objects in which the target event has occurred, Then go to step S404.

In step S404, the length of the risk indicator is gradually increased, thereby obtaining representative event characteristics of the prospecting target object that has occurred the target event, and then proceeding to step S408.

In step S405 , for the prospecting target objects without target events, the predetermined time interval is used as a sliding window to sample the respective events of the prospecting target objects without target events, and then proceed to step S406 .

In step S406, for the sampled events of the prospecting target objects without the target event, the risk index with a length of 1 and the corresponding time interval are used for prospecting to obtain the effective event characteristics of the prospecting target objects without the target event, Then proceed to step S407.

In step S407, the length of the risk indicator is gradually increased, thereby obtaining representative event characteristics of the surveying target objects without target events, and then proceeding to step S408.

In step S408, comparing the representative event characteristics of the prospecting target objects that have occurred the target event with the representative event characteristics of the prospecting target objects that have not occurred the target event, to obtain the prospecting target objects with potential event hotspots, and Assess the risk of target events occurring in prospecting targets with potential event hotspots.

A specific embodiment is listed below, which is the application of the equipment (such as server) and method for potential event hotspot detection of the present invention in road traffic safety.

First, the complex prospecting target object is a road data set, which is composed of n road data, expressed as R _i , i=1,...,n, each road data includes the name of the county, city, town and road, as shown in Table one shown.

Secondly, the respective characteristics of the plurality of exploration target objects are a road characteristic data set, which is composed of a plurality of road characteristic sets, as shown in Table 1.

C_i。這些道路特性項目共有r種，表示為I_i，i=1,...,r。第i種道路特性項目之道路特性項目所屬類別共有s種，表示為c_ij，j=1,...s,，並以集合表示為C_i={c_i1,c_i2,...,c_is}。每一道路特性集合係由r筆道路特性組成，表示為{(I₁,C₁’),(I₂,C₂’),...,(I_r,C_r’)}。 Each road characteristic is composed of road characteristic items and categories of road characteristic items, expressed as (I _i ,C _i '), C _i '

C _i . There are r types of these road characteristic items, expressed as I _i , where i=1,...,r. There are s types of road characteristic items of the i-th road characteristic item, expressed as c _ij , j=1,...s, and expressed as a set of C _i ={c _i1 ,c _i2 ,..., c _is }. Each road characteristic set is composed of r road characteristics, expressed as {(I ₁ ,C ₁ '),(I ₂ ,C ₂ '),...,(I _r ,C _r ')}.

These road characteristic items at least include road type, road line number, and road indication signs, and these road characteristic items are described as follows:

The type of road is expressed as I ₁ , and there are 3 types of it, and the set is expressed as C ₁ = {national road, express road, provincial and county road}. In order to facilitate subsequent grouping, numerical coding is carried out for this set, and after sequential coding A set is denoted as C ₁ ={1,2,3}. The number of road lines is expressed as I ₂ , which belongs to 4 categories, and is expressed as a set of C ₂ ={four-lane road, three-lane road, second-lane road, and one-lane road}. In order to facilitate subsequent grouping, the set is digitized Encoding, the set after sequential encoding is expressed as C ₂ ={1,2,3,4}. The road sign is expressed as I ₃ , which belongs to 4 categories, and is expressed as a set of C ₃ ={none, bumpy road, school, hospital}. In order to facilitate subsequent grouping, the set is numerically coded and coded sequentially The back set is denoted as C ₃ ={1,2,3,4}. For example, the set of road characteristics of road data R1 is {(I ₁ ,1),(I ₂ ,1),(I ₃ ,2)}, which means that the road characteristics of this road are: the road type is national road, and the number of road lines is Four-lane roads and road signs are road bumps.

Thirdly, the events of the multiple prospecting target objects are a road event data set, which is composed of multiple road event sets, as shown in Table 1.

O，E’

E，N’

N。車機編號共有3筆，並以集合表示為O={OBU1,OBU2,OBU3}。車機回報之危險指標共有6種，並以集合表示為E={超速,急加速,急減速,急過彎,前方防碰撞,車道偏移}，為便於後續事件特徵之探勘，針對該集合進行類別化編碼，依序編碼後集合表示為E={a,b,c,d,e}。回報時間共有t筆，表示為T_n，n=1,...,t。事故註記共有2種，並以集合表示為N={已發生事故,未發生事故}，為便於後續事件特徵之探勘，針對該集合進行類別化編碼，依序編碼後集合表示為N={T,F}。每一道路事件 集合係由u筆道路事件組成，表示為{(O’,E’,T₁,N’),(O’,E’,T₂,N’),...,(O’,E’,T_u,N’)}。 Each road event is composed of the vehicle number, the risk indicator reported by the vehicle, the return time and the accident note, expressed as (O',E',T _n ,N'), O'

O, E'

E, N'

N. There are 3 vehicle numbers, and they are expressed as O={OBU1,OBU2,OBU3} in aggregate. There are 6 types of risk indicators reported by vehicles, and they are expressed as a set of E={speeding, rapid acceleration, rapid deceleration, sharp cornering, front collision avoidance, lane deviation}. Carry out categorical encoding, and the set after sequential encoding is expressed as E={a,b,c,d,e}. There are t times of return time, denoted as T _n , n=1,...,t. There are 2 types of accident notes, and they are represented by a set as N={Accidents have occurred, No accidents}. In order to facilitate the exploration of the characteristics of subsequent events, the set is classified and coded, and the set is represented as N={T after sequential coding ,F}. Each road event set is composed of u road events, expressed as {(O',E',T ₁ ,N'),(O',E',T ₂ ,N'),...,(O ',E',T _u ,N')}.

Table 1: Road dataset.

Then, according to the characteristics of the complex prospecting target objects, the complex prospecting target objects are grouped.

In one embodiment, the clustering method is, for example, k-medoids in the divisional clustering method, which includes four steps:

In the first step, according to the number of clusters to be grouped, randomly select k prospecting target objects in the prospecting target data set as the center point; in the second step, calculate the relationship between these prospecting target objects and k The distance between the center points is divided into the same group with the nearest center point; in the third step, the distance between these exploration target objects is calculated according to the characteristic values of the exploration target objects in each group, and all The prospecting target object with the minimum distance and minimum is taken as the new center point; and in the fourth step, the old center point is replaced by the new center point according to the third step, and the above-mentioned second and third steps are repeated until the center point does not change any more.

The distance calculation method in the clustering method is, for example, the Euclidean distance calculation method. Two prospecting targets can be selected, and the distance between the two prospecting targets can be calculated according to the characteristic values of the two prospecting targets. The distance D can be calculated using the following formula:

In other words, the above formula is used to calculate the distance of the road data according to the categories of the road characteristic items as the basis for the similarity between the road data.

In the first step, the road data R ₁ ~ R ₇ are divided into two groups, and the two road data R ₁ and R ₅ are randomly selected as the center point; in the second step, the road data R is calculated according to the category of the road characteristic item. ₁ ~ R ₇ and the distance between two central points R ₁ and R ₅ , and they are divided into the same group as the nearest central point. Taking the road data R ₂ as an example, the road characteristic set {(I ₁ ,1),(I ₂ ,3),(I ₃ , ₁ )} of the road data R ₂ and the road characteristic set {( I ₁ ,1),(I ₂ ,1),(I ₃ ,2)} belong to the categories of road characteristic items, input the Euclidean distance calculation method, the distance between two points can be obtained as 2.236067977, and the road data R ₂ The road characteristic set {(I ₁ ,1),(I ₂ ,3),(I ₃ , ₁ )} and the road characteristic set {(I ₁ ,3),(I ₂ ,4), (I ₃ ,4)} The categories of road characteristic items belong to, input the Euclidean distance calculation method, and the distance between two points can be obtained as 3.741657387.

By analogy, the distance (similarity) between the road data R ₂ ~R ₄ , R ₆ ~R ₇ and the two center points R ₁ , R ₅ can be obtained, as shown in Table 2.

Table 2: Road data and the distance from the center point.

In Table 2, the distance between road data R ₂ and center point R ₁ is the shortest, so they are divided into the same group, and so on, the result of the first grouping is: Group 1 takes road data R ₁ as the center point, according to the closest Grouping by distance, the grouping result belongs to the road data of group 1 including R ₁ , R ₂ , R ₃ ; group 2 takes the road data R ₅ as the center point, and performs grouping according to the shortest distance, and the grouping result belongs to the road data of group 2 Contains R ₄ , R ₅ , R ₆ , R ₇ .

In the third step, the distance between the road data is calculated according to the category of the road characteristic items of the road data grouped in the grouping result of the second step, and the road data with the minimum sum of all distances is selected as the new center point.

Taking the road data R ₁ in group ₁ as an example, the road characteristic set {(I ₁ ,1),(I ₂ ,1),(I ₃ ,2)} of the road data R 1 and the road data R ₂ The category of the road characteristic items in the road characteristic set {(I ₁ ,1),(I ₂ ,3),(I ₃ ,1)} belongs to the category, input the Euclidean distance calculation method, the distance between two points can be obtained as 2.236067977, and then Road characteristic set {(I ₁ ,1), (I ₂ ,1), (I ₃ ,2)} of road data R ₁ and road characteristic set {(I ₁ ,1), (I ₂ , 2),(I ₃ ,2)} The category of the road characteristic item belongs to, input the Euclidean distance calculation method, the distance between two points is 1, and the distance sum is 3.236067977. By analogy, the sum of the distances between the road data in the groups can be obtained, as shown in Table 3 and Table 4.

Table 3: Distances between road data in Group 1.

Table 4: Distance between road data in group 2.

In Table 3, the road data R ₃ in group 1 has the smallest distance from other road data, so R ₃ is selected as the new center point; in Table 2, the distance between road data R ₆ in group 2 and other road data is the smallest. and the minimum, so choose R ₆ as the new center point.

In the fourth step, the old center point is replaced by the new center point according to the result of the third step, and the second and third steps are repeated until the center point does not change any more. The final grouping result is as follows: Group 1 uses road data _R The center point is grouped according to the shortest distance, and the grouping result belongs to the road data of group 1 including R ₁ , R ₂ , R ₃ ; the group 2 takes the road data R ₆ as the center point, and is grouped according to the shortest distance, and the grouping result belongs to The road data of group 2 includes R ₄ , R ₅ , R ₆ , and R ₇ .

Then, a plurality of representative event features are mined out for the events of the plurality of target objects in each group, including a plurality of representative event features mined from the events of the target objects where the target event has occurred, and from each A plurality of representative event characteristics detected by the event detection of the target object without the occurrence of the target event. In one embodiment, the target event is an accident, that is, according to the accident note of the road event in Table 1, it is distinguished whether each road data has an accident or not.

Taking group 2 as an example, the accident note of the road event in road data R ₆ contains T, so it is divided into road data that has occurred in accidents; and so on, the road data that group 2 distinguishes results in accidents that have occurred includes R ₆ , R ₇ , road data without accidents include R ₄ and R ₅ .

In addition, for the event of the target object of the target event that has occurred, the sampling method is the time interval before the occurrence of the target event. 5 minutes before it happened.

Taking the road data R6 in group ₂ as an example, first analyze whether the accident note of the road event is T, if it is T, it means that the road event is an accident occurrence point, after analysis, the road event belonging to the accident occurrence point includes ( OBU1, e, 0924, T), (OBU2, e, 0934, T), (OBU3, e, 0954, T).

According to the above analysis results, the road events from the point of accident occurrence to a period of time (for example, five minutes) before the occurrence of the accident are sampled respectively for these accident-prone vehicles to form a road event sequence. Taking the road event (OBU3,e,0954,T) as an example, the road event sequence after sampling is {(OBU3,b,0951,F), (OBU3,d,0952,F), (OBU3,c,0953, F), (OBU3,e,0954,T)}. By analogy, the road event sequences of the accident occurrence points in the group can be obtained. The sampling results are shown in Table 5.

Table 5: The road event sampling results of the road data of the accidents in group 2.

For the event data of the exploration target object that has not occurred the target event, the sampling method is a sliding window method, and one by one is sampled according to the time sequence. The window range of the sliding window method is a time interval, and the window translation distance of the sliding window method is is an adjustable parameter. For example, each road data captures the road events one by one according to the time sequence for the number of vehicles in the road event. The time interval of the window range of the sliding window method is 5 minutes, that is, the window translation distance is in time, and the unit is 5 minutes.

Taking the road data R4 in group ₂ as an example, the vehicle number of these road events is OBU1 for sampling. Sampled one by one from the whole point, the sliding window covers the time range from 12:00 to 12:05 for the first time, and the road events belonging to this time range include (OBU1, f, 1203, F), (OBU1, c, 1205, F). After the first shift of the sliding window, the second time covers the time range from 12:05 to 12:10, and no road event falls within this time range. After the sliding window is shifted for the second time, the third time covers the time range from 12:10 to 12:15. The road events within this time range include (OBU1,a,1211,F), (OBU1,c,1213, F), (OBU1, e, 1214, F), (OBU1, f, 1215, F). By analogy, the sampling results of each road data in the group can be obtained, as shown in Table 6.

Table 6: Sampling results of road incidents on road data without accidents in Group 2.

Then, according to the occurrence frequency, the relationship between occurrence time, the occurrence time interval and the scope of the event data of the exploration target object, the event characteristics are detected. The event feature mining method is a frequent episode mining analysis method, and the frequent episodes explored by this method include serial episodes, parallel episodes and composite episodes, where the serial episodes are expressed as There is a chronological sequence relationship between event features, parallel plots indicate that the time relationship between event features is not important, and compound plots are composed of ordered plots and parallel plots.

In one embodiment, the event feature mining method is, for example, a frequent scenario mining analysis method, which includes four steps:

In the first step, parameters related to exploration conditions are set. Except minimal-occurrences is used to avoid repeated calculation of event characteristics, other parameters related to exploration conditions include max-gap, min-gap, max-during, and min-sup; among them, max -gap indicates the maximum time unit of the occurrence time of two event features; min-gap indicates the minimum time unit of the occurrence time between two event features; max-during indicates the maximum time unit of the time interval covered by the event feature; min-sup indicates is the minimum support for the occurrence of event features; in the second step, scan the event data of the survey target object after sampling, obtain the set of event features with a length of 1 and their occurrence time intervals, and check whether they meet the min-sup , if it matches, put it into the effective event feature set; in the third step, take out an event feature from the effective event feature set as a prefix, and take out its occurrence time interval set, and then according to the time interval set, for the exploration target after sampling Scan the event data of the object to obtain the A set of event features and their occurrence time intervals are checked in order to see if they meet the relevant parameters of the exploration conditions; Repeat this step for the new valid event feature set until the new event feature set is empty; if there is no match, the event feature is regarded as a representative event feature and returned; in the fourth step, repeat The third step, until all the representative event features are obtained.

Using the above-mentioned exploration method, according to the road event sampling results of the road data that has occurred accidents and each road data that has not occurred accidents, a plurality of representative event characteristics composed of hazard indicators are explored.

Taking the road incident sampling results of the road accident data in Group 2 as an example, using the above-mentioned exploration method, a plurality of representative event characteristics composed of hazard indicators are explored.

In the first step, the parameters related to exploration conditions are set. Except minimal-occurrences is used to avoid repeated calculation of event characteristics, the other parameters related to exploration conditions are 1 minute for min-gap system, 5 minutes for max-gap system, and 5 minutes for max-duiing system. 5 minutes, min-sup is 0.5; in the second step, scan the sampling results of these road incidents, obtain a set of risk indicators with a length of 1 and their occurrence time intervals, and check in order whether they meet min-sup, if they meet , put into the valid event feature set.

Take the road event sampling results of road accidents that have occurred as an example, that is, R ₆ and R ₇ in Table 5, to obtain a set of risk indicators with a length of 1 and their occurrence time intervals, as shown in Table 7.

Table 7: Exploration information of hazard indicators whose length is 1 in the road event sampling results of the road data that have had accidents in Group 2.

In Table 7, the min-sup of risk indicator a is 3/6>=0.5, so it is included in the effective event feature set; the min-sup of risk indicator d is 2/6<0.5, so it is not included in the effective event feature set. By analogy, check whether min-sup is met in sequence, and after checking, the effective event feature set includes a, b, c, e, f.

In the third step, a dangerous indicator is taken out from the effective event feature set as a prefix, and its occurrence time interval set is taken out, and then according to the time interval set, the sampling results of these road events are scanned to obtain the risk index as a prefix. A set of dangerous indicators and their occurrence time intervals is used to check whether the relevant parameters of the exploration conditions are met in sequence; The feature set repeats this step until the new feature set with events is empty; if not, the risk indicator is regarded as a representative event feature and returned.

Take the danger index a as an example, take the danger index a as a prefix from the effective event feature set, scan the road event sampling results of the road data that have occurred accidents, and obtain the danger index with the danger index a as the prefix and the set of its occurrence time interval, As shown in Table 8. Prefixing hazard index a has no matching time interval for hazard index b. Also, Table 8 is prefixed with risk indicator a, and can be sampled with risk indicators b, c, e, f in addition, such as (b, a), (b, c), (b, e), (b, f ), (c,a), (c,b), (c,e), (c,f), (f,a), (f,b), (f,c), (f,e).

Table 8: The road event sampling results of road accidents in group 2. The hazard index exploration information prefixed with a.

In Table 8, the time units of the time intervals of the risk indicators (a, e) are 1, 2, and 2 respectively, all of which meet the limits of max-gap, min-gap, and max-during, and min-sup is 3/ 6>=0.5, so put in a new effective event feature set. By analogy, check whether the relevant parameters of the exploration conditions are met in sequence, then the new effective event feature set contains (a, e), and for this new event feature set, take out the danger indicator (a, e) as a prefix, for Sampling results of road accidents that have occurred, and obtain the set of hazard indicators prefixed with hazard indicators (a, e) and their occurrence time intervals. After exploration, there are no hazard indicators prefixed with hazard indicators (a, e) and their occurrence time intervals Therefore, the hazard indicators (a, e) are regarded as representative event characteristics and returned. By analogy, repeat this step until the new event feature set is empty.

In the fourth step, the third step is repeated until all representative event features are obtained.

Take the road incident sampling results of the road accident data in Group 2 as an example, that is, Table 7. The representative event feature set includes (a, e), (b, e), as shown in Table 9.

Table 9: Representative event feature set of the road data of the accidents in group 2.

By analogy, the representative event feature set of the road event sampling results of each accident-free road data in Group 2 can be obtained, as shown in Table 10.

Table 10: Representative event feature set of road data without accidents in Group 2.

Then, for the plurality of representative event features of each group, perform an event feature comparison to find out the prospecting target objects with potential event hotspots, and calculate an event feature for these prospecting target objects with potential event hotspots The degree of support is used to evaluate the frequency of occurrence of target events for those prospecting targets with potential event hotspots. The event feature comparison method is a complete comparison method, for example, a plurality of representative event features are extracted from the event data of each target object in each group that has not occurred the target event, if these representative If there is any event feature that is exactly the same as a plurality of representative event features mined from the event data of the prospecting target object that has occurred the target event, then the prospecting target object is regarded as a prospecting target object with potential event hotspots.

In other words, a plurality of representative event features are detected for each road event of road data without accidents in each group, if there is any event feature in these representative event features The characteristics of multiple representative events extracted from the road event data are exactly the same, indicating that these road data without accidents are potential accident hotspots, and the event feature support degree of these potential accident hotspots can be further calculated. Assess the likelihood of accidents occurring.

Taking group 2 as an example, set an accident occurrence threshold value such as 0.5, and the representative event characteristics of the road event sampling results of the road data R ₅ without accidents include (a, c), (a, e) These event features have any representative event features (a, e), (b, e) of the road event sampling results of the road data that have occurred accidents, and (a, e) are exactly the same, Therefore, the road data R ₅ is a potential accident hotspot; further calculation of its event feature support degree is 0.5, and its event feature support degree >=0.5, so it belongs to a high-prone potential accident hotspot.

By analogy, the road data comparison results of the accident-free roads in Group 2 can be obtained, as shown in Table 11.

Table 11: Comparison results of event characteristics of road data without accidents in Group 2.

Then, alerts are issued for the target events that have occurred and the prospecting target objects with potential event hotspots. The server stores the corresponding relationship between the road data of the accidents and potential accident hotspots and the adjacent base stations, and sends the warning information to the vehicle by analyzing the identification code of the base station reported by the vehicle.

Taking group 2 as an example, the corresponding relationship between the road data (such as R ₆ and R ₇ ) of accidents and the road data of potential accident hotspots (such as R ₅ ) and adjacent base stations is shown in Table 12.

Table 12: Correspondence table between road data requiring early warning and base station identification codes in Group 2.

To sum up, the equipment for potential event hotspot detection and its implementation method and computer program product of the present invention are to find out the prospecting targets with potential event hotspots by means of data collection, data grouping, event detection, event feature comparison, etc. , so that in traffic safety can effectively find the unoccurring Prospecting target objects with potential event hotspots for target events, and providing early warning to effectively reduce the frequency of target events.

The above-mentioned embodiments are only illustrative of the effects of this case, and are not intended to limit this case. Any person familiar with this technology can modify and change the above-mentioned implementations without violating the spirit and scope of this case. Therefore, the scope of protection of rights in this case should be listed in the scope of patent application described later.

S201~S205: steps

Claims (9)

A device for potential event hotspot exploration, including: a data collection module for exploration targets, which collects multiple exploration targets, their respective characteristics, and their respective events; a grouping module, According to the respective characteristics of the plurality of exploration target objects, the plurality of exploration target objects are grouped; the event feature detection module is based on whether a target event has occurred in each of the plurality of exploration target objects. Partitioning the prospecting target objects, wherein, for the prospecting target objects where the target event has occurred, using a predetermined time interval before the target event occurs as a window, sampling the respective events of the prospecting target objects where the target event has occurred; and For the prospecting target object without the target event, the predetermined time interval is used as the sliding window to sample the respective events of the prospecting target object without the target event. The event feature exploration module is based on the risk index and the occurrence time, Prospecting the exploration target object in each area in each group, and then obtaining the representative event characteristics of the exploration target object in each area in each group, wherein, for the target event that has occurred The sampled events of the exploration target object are explored with the risk index of length 1 and its corresponding time interval to obtain the effective event characteristics of the exploration target object that has occurred the target event, and then gradually increase the length of the risk index to obtain The representative event characteristics of the exploration target object where the target event has occurred; and for the sampled event of the exploration target object where the target event has not occurred, the risk index with a length of 1 and its corresponding time interval are used for exploration, Obtain the respective effective event characteristics of the prospecting target objects that have not occurred the target event, and then gradually increase the length of the risk index to obtain the respective representative event characteristics of the prospecting target objects that have not occurred the target event; and The event feature comparison module compares the representative event features of the prospecting targets in each region in each group to obtain prospecting targets with potential event hotspots. The device according to claim 1, wherein the grouping module selects a plurality of prospecting objects from the plurality of prospecting objects as center points, and calculates the unselected prospecting objects according to the characteristics of the plurality of prospecting objects Respective distances between the plurality of center points, so as to divide the unselected prospecting target object and the nearest center point into the same group, and then calculate the relationship between each prospecting target object in the same group and other survey target objects in the same group For the sum of distances between prospecting target objects, select the prospecting target object with the smallest distance sum as the new center point in the same group, and then recalculate with the new center point until the new center point does not change. The device according to claim 1, wherein the event feature comparison module evaluates the risk of the target event occurring on the prospecting target object with potential event hotspots for the prospecting target object with potential event hotspots. The device as described in Claim 1, wherein the plurality of prospecting objects are plural road data, the respective characteristics of the plurality of prospecting objects are the respective road characteristics of the plurality of road data, and the plurality of prospecting objects are Each event system includes a set consisting of vehicle serial number, danger indicator, return time, and whether an accident occurs. The device as described in claim 4 further includes: an event hotspot early warning module, which sends an early warning to the vehicle according to the base station identification code corresponding to the road data of the hot spots of the accidents that have occurred and potential accidents. A method for potential event hotspot detection performed by a server, comprising: collecting a plurality of detection objects, respective characteristics of the plurality of detection objects, the plurality of detection According to the respective events of the plurality of exploration target objects; according to the respective characteristics of the plurality of exploration target objects, group the plurality of exploration target objects; Partitioning the prospecting target objects, wherein, for the prospecting target objects where the target events have occurred, the respective events of the prospecting target objects where the target events have occurred are sampled using a predetermined time interval before the target event occurs as a window; And for the prospecting target objects that have not occurred target events, use the predetermined time interval as a sliding window to sample the respective events of the prospecting target objects that have not occurred target events; Prospecting for the exploration target object in each area in each group, to obtain the representative event characteristics of the exploration target object in each area in each group, wherein, for the exploration target object that has occurred in the target event, the sampled Events, the risk index with a length of 1 and its corresponding time interval are used for exploration to obtain the effective event characteristics of the exploration target object of the target event that has occurred, and then gradually increase the length of the risk index to obtain the exploration of the target event that has occurred The representative event characteristics of the target object; and for the sampled events of the exploration target object where the target event has not occurred, the risk index with a length of 1 and its corresponding time interval are used for exploration to obtain the Exploring the respective effective event characteristics of the target object, and then gradually increasing the length of the risk indicator, to obtain the respective representative event characteristics of the exploration target object that has not occurred the target event; The representative event characteristics of the exploration target objects in the area are compared with each other to obtain the exploration target objects with potential event hotspots. The method as described in claim 6, wherein the grouping method includes: (1) selecting a plurality of prospecting objects from the plurality of prospecting objects as center points; (2) selecting multiple prospecting objects according to the respective feature, calculate the respective distances between the unselected prospecting target object and the plurality of center points, so as to divide the unselected prospecting target object and the nearest center point into the same group; (3) calculate the same group The sum of distances between each prospecting target object and other prospecting target objects in the same group, so as to select the prospecting target object with the smallest distance sum as the new center point in the same group; and (4) use the new center Repeat steps (2) and (3) until the new center point does not change. The method as described in claim 6, wherein the comparison method includes: for the prospecting target object with potential event hotspots, assessing the risk of a target event occurring at the prospecting target object with potential event hotspots. A computer program product, which executes the method described in any one of claims 6-8 after the program is loaded into the computer.

Images