A design method for cluster risk control model based on big data

Abstract

The present invention discloses a design method for a cluster risk control model based on big data, the method comprising: step 1, selecting a monitoring location and designing video points, and deploying AI counting cameras; step 2, using a time series model to predict the total number of people in the cluster area by time period; step 3, using a police force deployment model to reasonably deploy police force according to the distribution of the number of people in each area by time period. Step 4, designing a patrol post area division model, automatically arranging and dispatching patrol robots, and timely discovering suspicious risks. Step 5, designing a risk control model, that is, establishing a cluster risk assessment system based on historical risk data, calculating the crowd gathering risk control situation index and predicting and warning in advance; achieving accurate prediction of abnormal aggregation, realizing optimal deployment of police force, and realizing adaptive patrol scheduling.

Description

Technical Field

[0001] The present invention relates to the research fields of big data, data mining and social governance, and in particular to a design method for a cluster risk control model based on big data. Background Art

[0002] In the process of realizing the present invention, the inventors found that there are at least the following problems in the prior art: the crowd concentration in open public places in cities is high, the pedestrian mobility is large, the risk uncertainty is large compared to closed places, and the probability of safety accidents such as crowding and trampling is very high. To control the risk of crowd gathering in a place, it is a more appropriate means to deploy police force. If there is no more reasonable police force deployment plan, there will be problems such as insufficient police force, excess police force, untimely police force dispatch, waste of equipment resources and low deployment efficiency in most densely populated open public places. These problems will directly increase the probability of various safety accidents and ultimately cause irreversible consequences. Therefore, it is necessary to design a cluster risk control model.

[0003] In order to overcome the shortcomings of the existing technology, the present invention provides a design method for a cluster risk control model based on big data, which can accurately predict abnormal clusters, optimize the deployment of police forces, and realize adaptive patrol scheduling. The technical solution is as follows:

[0004] The present invention provides a design method for a cluster risk control model based on big data, the method comprising the following steps:

[0005] Step 1: Select the monitoring location, design the video points, and deploy AI counting cameras;

[0006] The ICP algorithm is used to train the model to obtain an AI counting camera. Within the monitoring range, the AI ​​camera can analyze the three-dimensional information of people within the specified range and identify individual people, thereby realizing people counting.

[0007] Step 2: Use the time series model to predict the total number of people in the gathering area by time period;

[0008] After the AI ​​cameras are deployed and run for several days, the historical passenger flow patterns of each area are obtained, that is, the distribution of the number of people in each area in different time periods is obtained. According to the historical passenger flow patterns, the time series model is used to calculate the predicted value of the total number of people in each gathering area on the day by time period.

[0009] Step 3: Based on the population distribution in each area and time period, use the police force deployment model to reasonably deploy police force.

[0010] Police force deployment model: Set the number of regions to m, the total number of people in a certain region at a certain time period to p, the area of ​​the region to s, and the population density of the region to The area with population density f>3.8 is the crowd concentration point, and police force must be deployed. Multiple crowd concentration points form a crowd concentration point list. The distance between area i and area j is d ij ,i,j∈{1, 2, 3, ...m}, the number of patrol officers is n, and the dispatch speed is v. Assuming that when a suspicious risk event occurs in an area, the police must arrive within 10 minutes to avoid the incident; assuming that only one area has the risk of incident at the same time, the objective function is set as:

[0011] The constraints are:

[0012] in, L is a list of concentrated points of human traffic.

[0013] The above optimization problem can be solved using the solver cbc. According to the calculation results, multiple police force distribution points can be obtained and the police force deployment can be arranged in advance.

[0014] At the same time, the AI ​​counting camera counts the total number of people in the current area in real time ′ , update the list of concentrated crowd points every hour and update the police deployment model accordingly, and rearrange the police force according to the real-time results to achieve the patrol effect.

[0015] Step 4: Design a patrol post area division model to automatically schedule patrol robots and detect suspicious risks in a timely manner.

[0016] The police force distribution points obtained according to the police force deployment model are denoted as X = {x 1 ,x 2 ,…,x po}, construct Thiessen polygons according to the Delaunay triangulation method.

[0017] According to the Thiessen polygon method, the Thiessen polygon area is obtained, which corresponds to the patrol area. A corresponding number of patrol robots are arranged to perform fixed-line patrols in each patrol area.

[0018] Step 5: Design a risk control model, that is, establish a cluster risk assessment system based on historical risk data, calculate the crowd gathering risk control situation index and make early predictions and warnings; the details are as follows:

[0019] First, obtain the historical regional cluster risk data; manually mark the risk warning level based on the historical cluster risk feature labels.

[0020] The decision tree CART training model is used, with the independent variable being the risk feature and the dependent variable being the warning level, to extract the most relevant feature X=(x 11 ,…,x 21 ,…,x m′n,), corresponding weight A = (a 11 ,…,a 21 ,…,a m,n, ) and predicted probability Y = (y 1 ,y 2 ,…,y m′ ), the corresponding expression is as follows:

[0021]

[0022] Where m' is the number of levels and n' is the number of features.

[0023] Use the trained decision tree model for prediction: Input the historical behavior characteristics of the region into the trained decision tree model and return the warning level and risk probability y i ,i=1,…,m ′ .

[0024] Calculate the crowd gathering risk control situation index R i :

[0025]

[0026] Where R G k is the crowd gathering risk index; 1 The number of police officers deployed; k 2 is the number of patrol robots in step 4; y i is the risk probability. According to the crowd gathering risk control situation index R i Different police forces and measures are given according to the size of the area.

[0027] Preferably, in step 1, AI counting cameras are deployed at all entrances and exits of places where gatherings are prone to occur, so as to accurately obtain the total number of people in each area.

[0028] Preferably, in step 2, the predicted value of the total number of people in each gathering area on the day is calculated by time period, as follows:

[0029] According to the distribution of human flow in the history, the Fisher optimal segmentation method is used to divide the whole day into I time periods, so that the fluctuation or variance of human flow in each time period is minimized. For each time period, the time series model is used to predict the total number of people in each area on that day.

[0030] Preferably, the premise of the objective function in step 3 is: assuming that there is only one area with a risk of case occurrence at the same time, if there may be two areas with a risk of case occurrence at the same time, the constraint condition of the model is adjusted to:

[0031] Preferably, the step 4 of constructing Thiessen polygons is as follows:

[0032] Construct an initial triangle between the police force distribution points, and then draw a vertical line through the midpoint of each side to connect them to form a Thiessen polygon. The Thiessen polygon area is T(x i ), then T(x i )={x∈T(x i )|d(x,x i )≤d(x,x j ),j=1,…,po,j≠i}, where d represents the distance between two points.

[0033] Furthermore, step 4 also includes: timely broadcasting and uploading any abnormality identified during the patrol process, manually checking the broadcast video to confirm whether the alarm is triggered, and if the alarm is triggered, automatically linking nearby ordinary controllable cameras connected to the system to capture the on-site video at the same time, and automatically saving the video clip that triggered the alarm.

[0034] Preferably, the aggregate risk data described in step 5 specifically include the number of casualties, financial losses, number of onlookers, special groups involved, number of regional police officers when the risk occurs, total number of regional risks in the past year, and total number of risks in the past month.

[0035] Preferably, step 5 is based on the crowd gathering risk control situation index R i Different police forces and measures are given according to the size of the i ≥1, the regional risk is Class A warning, and the measures tend to increase a large number of police forces to immediately evacuate the crowd; 0.8≤R i <1 for regional risk, a Class B warning is issued, and measures tend to increase the number of police officers to guide the crowd and control the number of people; 0.6≤R i If the regional risk is less than 0.8, a Class C warning will be issued, and measures tend to be to increase a small number of police forces to guide the crowd; otherwise, no warning will be issued.

[0036] Furthermore, the crowd gathering risk index R in step 5 G for: Where f is the population density of the area; v′ is the average moving speed of the crowd; and h is the regional accessibility index (h∈[1,2]).

[0037] Preferably, step 5 also includes: adding the risk probability of each area obtained in step 5 to the police force deployment model in step 3 to ensure that more police forces can be deployed in areas with high risks.

[0038] Compared with the existing technology, one of the above technical solutions has the following beneficial effects: by deploying AI counting cameras, the time series model is used to predict the total number of people in the gathering area in different time periods, and abnormal gatherings are accurately predicted. According to the distribution of the number of people in each area in different time periods, the police force deployment model is used to reasonably deploy the police force to achieve the optimal deployment of the police force. By designing a patrol post area division model, patrol robots are automatically arranged and dispatched, suspicious risks are discovered in time, and adaptive patrol dispatch is achieved. At the same time, a risk control model is designed, and an aggregation risk assessment system is established based on historical risk data, and the crowd gathering risk control situation index is calculated to achieve early prediction, warning and prevention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A schematic diagram of a Thiessen polygon provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0040] In order to illustrate the technical solution and working principle of the present invention, the following will further describe the embodiments of the present invention in detail with reference to the accompanying drawings. All the above optional technical solutions can be combined in any way to form optional embodiments of the present invention, which will not be described one by one here.

[0041] The terms "step 1", "step 2", "step 3", and the like (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those described herein.

[0042] The present disclosure provides a method for designing a cluster risk control model based on big data, the method comprising the following steps:

[0043] Step 1: Select the monitoring location, design the video points, and deploy AI counting cameras;

[0044] The ICP algorithm is used to train the model to obtain an AI counting camera. Within the monitoring range, the AI ​​camera can analyze the three-dimensional information of people within the specified range and identify individual people, thereby realizing people counting.

[0045] Through field research, we can obtain all places with frequent gatherings in the area, such as vegetable markets, shopping malls, squares, schools, etc. Preferably, AI counting cameras are deployed at all entrances and exits of places prone to gatherings to accurately obtain the total number of people in each area, so as to facilitate further accurate detection of abnormal behaviors such as crowd gathering.

[0046] Step 2: Use the time series model to predict the total number of people in the gathering area by time period;

[0047] After the AI ​​cameras are deployed and run for several days, the historical passenger flow patterns of each area are obtained, that is, the distribution of the number of people in each area in different time periods is obtained. According to the historical passenger flow patterns, the time series model is used to calculate the predicted value of the total number of people in each gathering area on the day by time period.

[0048] Preferably, in step 2, the predicted value of the total number of people in each gathering area on the day is calculated by time period, as follows:

[0049] According to the distribution of human flow in the history, the Fisher optimal segmentation method is used to divide the whole day into I time periods, so that the fluctuation or variance of human flow in each time period is minimized. For each time period, the time series model is used to predict the total number of people in each area on that day.

[0050] Step 3: Based on the population distribution in each area and time period, use the police force deployment model to reasonably deploy police force.

[0051] According to existing research, the maximum safe density of a crowd is about 3.8 people / square meter. Assuming that the total number of people in a certain area during a certain period of time is p, the density of the population in the area is calculated based on the area s. When f>3.8, it is considered that there is a risk of case occurrence in the area.

[0052] Police force deployment model: Set the number of regions to m, the total number of people in a certain region at a certain time period to p, the area of ​​the region to s, and the population density of the region to The area with population density f>3.8 is the crowd concentration point, and police force must be deployed. Multiple crowd concentration points form a crowd concentration point list. The distance between area i and area j is d ij ,i,j∈{1, 2, 3,...m}(d ij The unit is meter), the number of patrol officers is n, and the dispatch speed is v (v is meter / minute, assuming that everyone's dispatch speed is the same). Assume that when a suspicious risk event occurs in an area, the police must arrive within 10 minutes to avoid the incident; in order to deploy the least police force and ensure that all areas can avoid incidents, it is assumed that only one area has the risk of incidents at the same time, and the objective function is set as:

[0053] The constraints are:

[0054] in, L is a list of concentrated points of human traffic.

[0055] The above optimization problem can be solved using the solver cbc. According to the calculation results, multiple police force distribution points can be obtained and the police force deployment can be arranged in advance.

[0056] At the same time, the AI ​​counting camera counts the total number of people in the current area in real time ′, update the list of concentrated crowd points every hour and update the police deployment model accordingly, and rearrange the police force according to the real-time results to achieve the patrol effect.

[0057] Preferably, the above premise is that there is only one area with a risk of case occurrence at the same time. If there may be two areas with a risk of case occurrence at the same time, the constraint condition of the model is adjusted to:

[0058]

[0059] Step 4: Design a patrol post area division model to automatically schedule patrol robots and detect suspicious risks in a timely manner.

[0060] The police force distribution points obtained according to the police force deployment model are denoted as X = {x 1 ,x 2 ,…,x p}, construct Thiessen polygons according to the Delaunay triangulation method.

[0061] Preferably, the constructing of Thiessen polygons is specifically as follows:

[0062] Construct an initial triangle between the police force distribution points, then draw a vertical line through the midpoint of each side to connect them to form a Thiessen polygon. Figure 1 is a schematic diagram of a Thiessen polygon. The Thiessen polygon area is T(x i ), then T(x i )={x∈T(x i )|d(x,x i )≤d(x,x j ),j=1,…,p,j≠i}, where d represents the distance between two points.

[0063] According to the Thiessen polygon method, the Thiessen polygon area is obtained, which corresponds to the patrol area. A corresponding number of patrol robots are arranged to perform fixed-line patrols (i.e., patrols along the prescribed routes) in each patrol area.

[0064] Preferably, any abnormality identified during the patrol process is promptly broadcast and uploaded, and the broadcast video is manually checked to confirm whether the alarm is triggered. If the alarm is triggered, it will automatically link with nearby ordinary controllable cameras connected to the system to capture the live video at the same time, and automatically save the video clip that triggered the alarm.

[0065] Step 5: Design a risk control model, that is, establish an aggregation risk assessment system based on historical risk data, calculate the crowd aggregation risk control situation index and provide early prediction and warning.

[0066] The details are as follows:

[0067] First, obtain the historical regional cluster risk data; according to the historical cluster risk feature labels, manually mark the risk warning level (divided into 4 levels: A red warning, B orange warning, C yellow warning, and D normal).

[0068] Preferably, the aggregated risk data specifically include the number of casualties, financial losses, number of onlookers, and special groups involved; the number of regional police officers when the risk occurs; the total number of regional risks in the past year; and the total number of risks in the past month.

[0069] The decision tree (CART) training model is used, with the independent variable being the risk feature and the dependent variable being the warning level, to extract the most relevant feature X=(x 11 ,…,x 21 ,…,x m′n′ ), corresponding weight A = (a 11 ,…,a 21 ,…,a m′n′ ) and predicted probability Y = (y 1 ,y 2 ,…,y m′ ), the corresponding expression is as follows:

[0070]

[0071] Where m' is the number of levels and n' is the number of features.

[0072] Use the trained decision tree model for prediction: Input the historical behavior characteristics of the region into the trained decision tree model and return the warning level and risk probability y i ,i=1,…,m ′ .

[0073] Calculate the crowd gathering risk control situation index R i :

[0074]

[0075] Where R G k is the crowd gathering risk index; 1 The number of police officers deployed; k 2 is the number of patrol robots in step 4; y i is the risk probability. According to the crowd gathering risk control situation index R i Different police forces and measures are given according to the size of the area.

[0076] Preferably, step 5 is based on the crowd gathering risk control situation index R i Different police forces and measures are given according to the size of the i ≥1, the regional risk is Class A warning, and the measures tend to increase a large number of police forces to immediately evacuate the crowd; 0.8≤Ri <1 for regional risk, a Class B warning is issued, and measures tend to increase the number of police officers to guide the crowd and control the number of people; 0.6≤R i If the regional risk is less than 0.8, a Class C warning will be issued, and measures tend to be to increase a small number of police forces to guide the crowd; otherwise, no warning will be issued.

[0077] Preferably, the crowd gathering risk index R in step 5 G for: Where f is the population density of the area; v′ is the average moving speed of the crowd; and h is the regional accessibility index (h∈[1,2]).

[0078] It can be seen that the larger the crowd gathering risk index is, the more dangerous the crowd gathering situation is; the higher the regional accessibility index is, the more accessible the area is; the smaller the gathering risk index is, and the more police officers deployed are, the stronger the prevention capability of the area is.

[0079] Preferably, step 5 further includes: adding the risk probability of each area obtained in step 5 to the police force deployment model in step 3 to ensure that more police forces can be deployed in areas with high risks. Specifically, the population density of the area is multiplied by the risk coefficient w, that is, the population density of the area in y is the regional risk probability, eps=0.000001 (to avoid the denominator being 0).

[0080] The present invention is described above by way of example in conjunction with the accompanying drawings. Obviously, the specific implementation of the present invention is not limited to the above-mentioned method. All non-substantial improvements made by adopting the method concept and technical solution of the present invention; or directly applying the above-mentioned concept and technical solution of the present invention to other occasions without improvement or equivalent replacement, are within the protection scope of the present invention.

Claims

1. A design method for a cluster risk control model based on big data, characterized in that: The method comprises the following steps: Step 1: Select the monitoring location, design the video points, and deploy AI counting cameras; The ICP algorithm is used to train the model and obtain an AI counting camera. Within the monitoring range, the AI ​​camera can analyze the three-dimensional information of people within the specified range and identify the individual people, thereby realizing people counting. Step 2: Use the time series model to predict the total number of people in the gathering area by time period; After the AI ​​cameras are deployed and run for several days, the historical passenger flow patterns of each area are obtained to obtain the distribution of the number of people in each area by time period. Based on the historical passenger flow patterns, the time series model is used to calculate the predicted value of the total number of people in each gathering area on the day by time period; Step 3: Based on the distribution of people in each area and time period, use the police force deployment model to reasonably deploy police force; Police force deployment model: Set the number of regions to m, the total number of people in a certain region at a certain time period to p, the area of ​​the region to s, and the population density of the region to The area with population density f>3.8 is the crowd concentration point, and police force must be deployed. Multiple crowd concentration points form a crowd concentration point list. The distance between area i and area j is d ij ,i,j∈{1, 2, 3, ...m}, the number of patrol officers is n, the dispatching speed is v, and if a suspicious risk event occurs in an area, the police must arrive within 10 minutes to avoid the incident; if there is only one area with a risk of incident at the same time, the objective function is set as: The constraints are: in, L is a list of concentrated flow points; The optimization problem can be solved by using the solver cbc. According to the calculation results, multiple police force distribution points can be obtained, and the police force deployment can be arranged in advance. At the same time, the AI ​​counting camera counts the total number of people in the current area in real time ′ , update the list of crowd concentration points every hour and update the police deployment model accordingly, and rearrange the police force according to the real-time results to achieve the patrol effect; Step 4: Design a patrol post area division model to automatically schedule patrol robots and detect suspicious risks in a timely manner; The police force distribution points obtained according to the police force deployment model are denoted as X = {x1, x2, ..., x po }, construct Thiessen polygons according to the Delaunay triangulation method; According to the Thiessen polygon method, the Thiessen polygon area is obtained, which corresponds to the patrol area, and a corresponding number of patrol robots are arranged to perform fixed-line patrols in each patrol area; Step 5: Design a risk control model, establish a cluster risk assessment system based on historical risk data, calculate the crowd gathering risk control situation index and make early predictions and warnings; the details are as follows: First, obtain the historical regional cluster risk data; manually mark the risk warning level based on the historical cluster risk feature labels; The decision tree CART training model is used, with the independent variable being the risk feature and the dependent variable being the warning level, to extract the most relevant feature X=(x 11 ,…,x 21 ,…,x m′n′ ), corresponding weight A = (a 11 ,…,a 21 ,…,a m′n′ ) and predicted probability Y = (y1,y2,…,y m′ ), the corresponding expression is as follows: Where m' is the number of levels and n' is the number of features; Use the trained decision tree model for prediction: Input the historical behavior characteristics of the region into the trained decision tree model and return the warning level and risk probability y i ,i=1,…,m ′ ; Calculate the crowd gathering risk control situation index R i : Where R G is the crowd gathering risk index; k1 is the number of police officers deployed; k2 is the number of patrol robots in step 4; y i is the risk probability; eps = 0.000001; according to the crowd gathering risk control situation index R i Different police forces and measures are given according to the size of the area.

2. The design method of a cluster risk control model based on big data according to claim 1, characterized in that: In step 1, AI counting cameras are deployed at all entrances and exits of places where gatherings are likely to occur, so as to accurately obtain the total number of people in each area.

3. The design method of a cluster risk control model based on big data according to claim 1, characterized in that: In step 2, the predicted total number of people in each gathering area on the day is calculated by time period, as follows: According to the distribution of human flow in the history, the Fisher optimal segmentation method is used to divide the whole day into I time periods, so that the fluctuation or variance of human flow in each time period is minimized. For each time period, the time series model is used to predict the total number of people in each area on that day.

4. The design method of a cluster risk control model based on big data according to claim 1, characterized in that: The premise of the objective function in step 3 is: assuming that only one region has the risk of a case occurring at the same time. If there are two regions with the risk of a case occurring at the same time, the constraints of the model are adjusted to:

5. The design method of a cluster risk control model based on big data according to claim 1, characterized in that: Step 4 describes the construction of Thiessen polygons, specifically: Construct an initial triangle between the police force distribution points, and then draw a vertical line through the midpoint of each side to connect them to form a Thiessen polygon. The Thiessen polygon area is T(x i ), then T(x i )={x∈T(x i )|d(x,x i )≤d(x,x j ),j=1,…,po,j≠i}, where d represents the distance between two points.

6. The design method of a cluster risk control model based on big data according to claim 5 is characterized in that: Step 4 also includes: timely broadcasting and uploading any abnormality identified during the patrol process, manually checking the broadcast video to confirm whether the alarm is triggered, and if the alarm is triggered, automatically linking nearby ordinary controllable cameras connected to the system to capture the on-site video at the same time, and automatically saving the video clip that triggered the alarm.

7. A design method for a cluster risk control model based on big data according to any one of claims 1 to 6, characterized in that: The clustered risk data described in step 5 specifically include the number of casualties, financial losses, number of onlookers, special groups involved, number of regional police officers when the risk occurs, total number of regional risks in the past year, and total number of risks in the past month.

8. The design method of a cluster risk control model based on big data according to claim 7 is characterized in that: Step 5: Control the situation index R based on the crowd gathering risk i Different police forces and measures are given according to the size of the i A Class A warning is issued for areas with a risk of ≥1, and measures tend to increase police force and immediately evacuate people; 0.8≤R i Areas with a risk less than 1 will be given a Class B warning, and measures tend to include increasing police force to guide the crowd and control the number of people; 0.6≤R i If the regional risk is less than 0.8, a Class C warning will be issued, and measures tend to be to increase a small number of police forces to guide the crowd; otherwise, no warning will be issued.

9. The design method of a cluster risk control model based on big data according to claim 8, characterized in that: Crowd gathering risk index R in step 5 G for: Where f is the population density of the area; v′ is the average moving speed of the crowd; and h is the regional accessibility index.

10. A design method for a cluster risk control model based on big data according to any one of claims 8-9, characterized in that: Step 5 also includes: adding the risk probability of each area obtained in step 5 to the police force deployment model in step 3 to ensure that more police forces can be deployed in areas with high risks.

Images