CN110472775A - A kind of series case suspect's foothold prediction technique - Google Patents

Abstract

The invention discloses a method for predicting the whereabouts of suspects in a series of cases. The method first obtains the residence and activity range information of all mobile phone users, and generates the first residence kernel density map according to the residence information; The crime location information of the suspect is matched with the residence information of the mobile phone user, and a second residence kernel density map is generated; the crime distance attenuation is obtained by combining the crime location information of the suspect to be predicted with the crime location and residence information of the captured criminal function algorithm to obtain the probability surface distribution of the residence model of the suspect to be predicted; according to the first and second residence kernel density maps and the probability surface distribution of the residence model, combined with the Bayesian algorithm, obtain the Probability Density Plot. Adopting the technical scheme of the present invention does not need to always consider factors such as time, region, traffic environment, etc., so that the accuracy of the prediction model is reduced, and the accuracy of the result of predicting the suspect's foothold is improved.

Description

A method for predicting the whereabouts of suspects in a series of cases

technical field

The invention relates to the technical fields of public security and crime geography, in particular to a method for predicting the whereabouts of suspects in a series of cases.

Background technique

The series of cases have the characteristics of many criminals committing crimes, skilled crime methods, and strong anti-detection capabilities, which are likely to cause social panic and are not conducive to maintaining social stability. Therefore, being able to accurately predict the whereabouts of suspects in a series of cases can effectively improve the efficiency of police investigation work.

In the prior art, a method of constructing a Journey-to-crime estimation model (Journey-to-crime, JTC model for short) is usually used to predict the final destination of a suspect in a series of cases. The JTC model can construct a crime distance attenuation function from the series of crimes that have been solved, and then calculate the area that is most likely to be the place where the suspect in the series of cases ends. However, the current crime journey estimation model needs to always consider factors such as time, region, and traffic environment that affect the prediction accuracy of the JTC model. When the suspect commits a crime in the surrounding area of residence, predicting the residence information of the suspect according to the JTC model can get better prediction results; when the suspect commits a crime in an area far away from the residence, predicting the residence information of the suspect according to the JTC model , the prediction effect of the JTC model is not good. Therefore, the accuracy and accuracy of the existing JTC models still need to be improved.

Contents of the invention

The embodiment of the present invention proposes a method for predicting the whereabouts of suspects in a series of cases. It does not need to consider factors such as time, region, and traffic environment at all times to reduce the accuracy of the prediction model, and improves the accuracy of the results of predicting the whereabouts of suspects.

An embodiment of the present invention provides a method for predicting the whereabouts of suspects in a series of cases, including:

Obtaining first residence information and activity range information of all mobile phone users, wherein the first residence information is used to generate a first residence kernel density map of all mobile phone users;

According to the information on the scope of activities, according to the time and place, extract the mobile phone users that match the first crime location information of the series of cases to be predicted suspects, and according to the extracted first residence information corresponding to the mobile phone users, combined with the preset A spatial kernel density analysis algorithm to obtain a kernel density map of the second place of residence of the suspect to be predicted;

According to the preset information on the second place of crime and the second place of residence of the serial cases of the criminals captured, combined with the preset negative exponential function to fit the distance distribution of the second place of crime - the second place of residence Curve to obtain the crime distance decay function;

According to the information of the first crime location, combined with the crime distance attenuation function algorithm, the probability surface distribution of the residence model of the suspect to be predicted is obtained;

According to the first residential area kernel density map, the second residential area kernel density map and the probability surface distribution of the residential area model, combined with a preset Bayesian algorithm, obtain the location used to predict the suspect to be predicted. The probability density plot of .

Optionally, the acquisition of the first place of residence information and activity range information of all mobile phone users is specifically:

The first place of residence information includes several places of residence, and the activity range information includes several activity ranges and activity time corresponding to the activity ranges respectively;

According to the mobile phone signaling data of all mobile phone users, extract and aggregate to obtain all base station data;

According to all the base station point data, construct the base station Thiessen polygon, and combine the preset Monte Carlo simulation algorithm to obtain the base station positions corresponding to each mobile phone user at each moment;

Using the information entropy value method to calculate the positions of each of the base stations to obtain the first residence information and activity range information of all mobile phone users.

Optionally, the use of the information entropy value method to calculate the positions of each of the base stations is specifically:

Extract the base station positions corresponding to the night time period of each mobile phone user, and calculate the stay ratio of each mobile phone user at all base station positions during the night time period;

Calculate the total entropy value of each mobile phone user at night according to the corresponding stay ratio of all mobile phone users;

When the nighttime total entropy value of the i mobile phone user is less than the preset threshold, the base station position corresponding to the i mobile phone user with the largest staying ratio is marked as the residence of the i mobile phone user, and the i mobile phone user The location of the i-th mobile phone user other than the place of residence is marked as the activity range of the i-th mobile phone user.

Optionally, the spatial kernel density analysis algorithm uses a quartic kernel function.

Optionally, the second crime location information and the second residence information of the series of cases of the captured criminals are combined with a preset negative exponential function to fit the second crime location information - the second The distance distribution curve of the residence information obtains the crime distance decay function, specifically:

The information on the second place of crime includes several second places of crime, and the information on the second place of residence includes a second place of residence corresponding to the second place of crime respectively;

Calculating the distance between each of the second places of crime and its corresponding second place of residence to obtain crime distance data;

Taking the crime distance data as the horizontal axis and the case ratio as the vertical axis, the crime distance distribution curve is obtained;

Using a negative exponential function to fit the crime distance distribution curve, obtain the crime distance attenuation function f(d _ij ):

d _ij refers to the Euclidean distance to the residential point, and the parameters a _k , b _k , and c _k are constant values, which need to be obtained through the training of the crime distance distribution data of the actually captured criminals.

Optionally, according to the information of the first crime location, combined with the crime distance attenuation function algorithm, the probability surface distribution of the residence model of the suspect to be predicted is obtained, specifically:

The information of the first crime location includes several first crime locations and the first crime time corresponding to the first crime locations respectively;

The calculation method to obtain the probability surface distribution of the residence model of the suspect to be predicted is as follows:

Generate a grid of N*N meters in the prediction area, calculate the distance x _ij from each grid center point to the first crime site, and substitute the x _ij into the distance attenuation function, and calculate each Grids are used as the probability value of the residence model of the suspect's foothold to be predicted, and finally the probability surface distribution of the residence model of the suspect to be predicted is obtained.

Optionally, according to the first residential area kernel density map, the second residential area kernel density map and the probability surface distribution of the residential area model, combined with a preset Bayesian algorithm, the The probability density map for predicting the whereabouts of the suspect is calculated as follows:

Wherein, P(JTC|O) is the probability density of the to-be-predicted suspect’s foothold, P(JTC) is the probability surface distribution of the residence model of the to-be-predicted suspect, and P(O|JTC) is the probability density of the second residential area. Earth core density map, P(O) is the core density map of the first habitable place.

Implementing the embodiment of the present invention has the following beneficial effects:

The method for predicting the whereabouts of suspects in a series of cases provided by the embodiments of the present invention first obtains the residence and activity range information of all mobile phone users, and generates the first residence kernel density map according to the residence information; The location information of mobile phone users matched with the crime location information, and generate the second residence kernel density map; the crime distance attenuation function is obtained by combining the crime location information of the suspect to be predicted with the crime location and residence information of the captured criminal Algorithm to obtain the probability surface distribution of the residence model of the suspect to be predicted; according to the first and second residence kernel density maps and the probability surface distribution of the residence model, combined with the Bayesian algorithm, the probability used to predict the whereabouts of the suspect to be predicted is obtained Density map. Compared with the prior art that uses the traditional JTC model to predict the whereabouts of suspects, the technical solution of the present invention does not always consider the impact of factors such as time, region, and traffic environment on the JTC model’s prediction accuracy reduction, but will use the mobile phone signaling Data or big data with location information is used to study individual spatial behavior, which further improves the accuracy of the JTC model in predicting the results of the suspect's foothold.

Further, the present invention extracts and aggregates all base station data according to the mobile phone signaling data of mobile phone users, and processes the base station data in combination with Thiessen polygon and Monte Carlo simulation algorithms, which can effectively improve the mobile phone user's certain The data of the specific location in a certain base station at any time is more accurate.

Furthermore, when the present invention obtains the first residence information and activity range information of all mobile phone users, the entropy method is used to process the base station positions corresponding to each mobile phone user at each time, which further improves the acquisition of all mobile phone users. The accuracy of mobile phone users' primary residence information and activity range information.

Further, when acquiring the suspect to be predicted, the present invention extracts the mobile phone users that match the first crime location information of the series of cases of the suspect to be predicted according to the activity range information of all mobile phone users according to time and place, so that the suspect can be obtained. The data of the predicted suspect is not affected by time and region, which effectively improves the accuracy of obtaining the data of the suspect to be predicted.

Further, when the present invention obtains the probability density map of the suspect's foothold to be predicted, the Bayesian box algorithm is used to process the data, which improves the accuracy of the result of predicting the suspect's foothold.

Description of drawings

Fig. 1 is a schematic flow chart of an embodiment of a method for predicting the whereabouts of suspects in a series of cases provided by the present invention;

Fig. 2 is the flow chart of mobile phone signaling data processing in the present invention;

Fig. 3 is a schematic diagram of a base station and a Thiessen polygon in the present invention;

FIG. 4 is a fitting diagram of a distance attenuation curve in an embodiment of the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

Example 1

Please refer to FIG. 1 , which is a schematic flow chart of an embodiment of a method for predicting the whereabouts of suspects in a series of cases provided by the present invention. As shown in Figure 1, the construction method includes steps 101 to 105, and each step is specifically as follows:

Step 101: Obtain first residence information and activity range information of all mobile phone users, wherein the first residence information is used to generate a first residence kernel density map of all mobile phone users.

In this embodiment, the first residence information of all mobile phone users includes: several first residences; the activity range information of all mobile phone users includes: several activity ranges and activity time corresponding to the activity ranges respectively.

In this embodiment, since there are several base station signal receivers on the same latitude and longitude, there are also several groups of mobile phone signaling data on the same base signal receiver. Therefore, it is necessary to extract and aggregate the mobile phone signaling data of all mobile phone users, gather the base stations on the same longitude and latitude into one point, and obtain all the base station point data. This solution only extracts the mobile phone signaling data of mobile phone users under the premise of police investigation work, which is not an invention that violates 3.1.3 of the Patent Law and hinders public interests.

In this embodiment, step 101 is specifically: according to the mobile phone signaling data of all mobile phone users, extract and aggregate all base station point data; according to all base station point data, construct a base station Thiessen polygon, and combine the preset Monte Carlo The simulation algorithm obtains the base station positions corresponding to each mobile phone user at each moment; uses the information entropy value method to calculate the positions of each base station respectively, and obtains the first residence information and activity range information of all mobile phone users; according to the first The residence information is combined with the spatial kernel density analysis algorithm to generate the first residence kernel density map of all mobile phone users.

In this step, the base station positions corresponding to each mobile phone user at each moment are obtained, specifically: in the research area, when the i-th mobile phone user is located at j base station at time k, take j base station (X _j , Y _j ) as the center to construct a Thiessen polygon (see Figure 3), so that the Euclidean distance from any base point (x _p , y _p ) in the polygon to j base point (X _j , Y _j ) is smaller than the base point (x _p ,y _p ) to any other Euclidean base point. The Euclidean distance is defined as: The minimum enclosing rectangle is generated according to the Thiessen polygon centered on the base station j (X _j , Y _j ), and random points are generated by combining the Monte Carlo simulation algorithm. The probability that the i-th mobile phone user is located in the Thiessen polygon of base station j is set to 1, and the probability outside the Thiessen polygon and within the smallest enclosing rectangle is set to 0. Record the random point in the Thiessen polygon constructed by j base station as the specific base station position of the i-th mobile phone user at time k, and store the base station position corresponding to the i-th mobile phone user at time k through the mobile phone user ID. The above processing is performed on the base station where each user is located at each time, and the corresponding base station position information of each mobile phone user at each time is obtained. See Figure 2 for a detailed mobile phone signaling processing flow for mobile phone users.

In this step, use the information entropy value method to calculate the positions of each base station respectively, specifically: extract the base station positions corresponding to the night time period of each mobile phone user, and calculate the stay ratio of each mobile phone user at all base station positions in the night time period; Calculate the total nighttime entropy of each mobile phone user according to the corresponding stay ratio of all mobile phone users; The base station location is marked as the i-th mobile phone user's residence, and the i-th mobile phone user's other base station locations outside the residence are marked as the i-th mobile phone user's activity range.

For example, extract all base station locations where the i-th mobile phone user is at night time (such as 0-6 o'clock), and calculate the stay ratio of the i-th mobile phone user at all base station locations. The calculation method of the stay ratio is: Among them, P _ij represents the proportion of the i-th mobile phone user staying at the base station j, T _ij is the length of time the i-th mobile phone user stays at the j base station, and T is a constant, representing the total night time included in the calculation; then calculate the i-th The total entropy value of a mobile phone user at night is calculated as follows: Among them, H(U _i ) is the total entropy value of the i-th mobile phone user at night, P _ij represents the proportion of the i-th mobile phone user staying at the j base station location, n represents the i-th mobile phone user staying at n base station locations in total, and the entropy value The smaller H(U _i ), it means that the i-th mobile phone user switches between base station locations less frequently at night, and stays stably in a certain base station location.

When the night time period is set to 0-6 o’clock, 1.5 is used as the threshold, and if the H(U _i ) value is ≤1.5, the base station location with the largest proportion of the i-th mobile phone user staying in the night time period is selected as the i-th mobile phone user’s The place of residence, while the trajectory points generated by the i-th mobile phone user in other areas other than the place of residence are recorded as the activity range.

Step 102: According to the activity range information, according to the time and place, extract the mobile phone users that match the first crime location information of the series of cases to be predicted suspects, and combine the preset The spatial kernel density analysis algorithm is used to obtain the second residence kernel density map of the suspect to be predicted.

In this embodiment, the information on the first crime location of the series of cases of the suspect to be predicted includes several first crime locations and first crime times corresponding to each of the several first crime locations.

In this embodiment, step 102 is specifically: grid the urban space, obtain K square grids of N×N meters and add a number to each grid, and then obtain the data of each mobile phone user through spatial overlapping analysis. The grid number where the range of activities of the suspect is located, and the grid number where the first crime site of the series of cases of the suspect to be predicted is located. According to the grid number where the first crime site is located=the grid number where the mobile phone user's activity range is located, the first crime time=the mobile phone user's activity time, screen out the mobile phone users that match the serial cases of the suspect to be predicted. Next, extract the first place of residence corresponding to the mobile phone user that matches the series of cases of the suspect to be predicted, use the spatial kernel density analysis algorithm, and use the grid side length N meters as the search radius of the kernel density to generate the second place of residence of the suspect to be predicted. Habitat Kernel Density Map.

The spatial kernel density algorithm in this step uses a quartic kernel function. For the first place of residence (X _i , Y _i ) in the first place of residence information corresponding to the mobile phone user corresponding to the series of cases to be predicted suspects, any base station position (x _j , y _j ) within N meters around, The kernel function value at the base station position (x _j , y _j ) is obtained by the formula, Among them, k is a constant; t=d _ij /N, d _ij is the Euclidean distance between the base station location (x _j , y _j ) and the first residence (X _i , Y _i ), and N is the kernel density search radius. The spatial kernel density analysis algorithm adopted is a quartic kernel function, which can make the data distribution in the residential kernel density map more accurate.

Step 103: According to the preset information on the second place of crime and the information on the second place of residence of the series of cases of the captured criminals, combined with the preset negative exponential function to fit the distance between the information on the second place of crime and the information on the second place of residence distribution curve to obtain the crime distance decay function.

In this embodiment, step 103 is specifically: the information on the second place of crime includes several second places of crime, and the information on the second place of residence includes the second place of residence corresponding to the second places of crime respectively; The distance between the place where the crime was committed and its corresponding second place of residence is to obtain the crime distance data, and use the crime distance data to construct the crime distance distribution curve, with the segmented crime distance data as the horizontal axis, for example, 0-1000m, 1001- 2000m. The proportion of cases is on the vertical axis. For example, the vertical axis is the proportion of cases within the segmented crime distance interval. And choose the negative exponential function to fit the crime distance distribution curve, and output the crime distance decay function. The crime distance decay function f(d _ij ) is:

d _ij refers to the Euclidean distance between the base station location (x _j , y _j ) and the second place of residence (X _i , Y _i ), the parameters a _k , b _k , and c _k are constant values, and it is necessary to pass the actual captured criminals The crime distance distribution curve training is obtained.

In this step, the crime distance distribution curve is drawn by obtaining the data of the second crime location information and the second residence information information of a large number of serial cases of captured criminals, and a negative exponential function is used to fit the distribution curve, so that the construction crime distance attenuation The algorithm of the function is more accurate.

Step 104: According to the information of the first crime location, combined with the crime distance attenuation function algorithm, obtain the probability surface distribution of the residence model of the suspect to be predicted.

In this embodiment, the calculation method of step 104 is as follows: Generate a grid of 100m*100m in the prediction area, calculate the distance x _ij from the center point of each grid to the first crime site, and substitute x _ij into the crime distance attenuation function, and calculate each grid as the to-be-predicted The probability value of the residence model of the suspect's foothold, and finally obtain the probability surface distribution of the residence model of the suspect to be predicted.

Step 105: According to the first residential area kernel density map, the second residential area kernel density map and the probability surface distribution of the residential area model, combined with the preset Bayesian algorithm, obtain a probability density map for predicting the whereabouts of the suspect to be predicted.

In this embodiment, the calculation method of step 105 is as follows:

Among them, P(JTC|O) is the probability density of the suspect’s foothold to be predicted, P(JTC) is the probability surface distribution of the residence model of the suspect to be predicted, P(O|JTC) is the kernel density map of the second residence, and P (O) is the kernel density map of the first residence.

It can be seen from the above that by applying the technical solution of this embodiment, the residence, activity range and activity time of each mobile phone user can be obtained through the mobile phone signaling data. According to the location and time of the crime, the activity range and activity time of each mobile phone user extracted based on the mobile phone signaling data are matched to the suspect to be predicted. The acquisition of the data of the suspect to be predicted can be prevented from being affected by time and region, and the accuracy of obtaining the data of the suspect to be predicted can be effectively improved.

In order to better illustrate the process and principle of this embodiment, the following example is used as a detailed description:

Obtain 10,785,815 pieces of mobile phone signaling data of all mobile phone users in ZG City on December 28, 2018. Through the extraction and aggregation of mobile phone signaling data, 10,594 base station data are obtained, and then Thiessen polygons are constructed with each base station as the center. According to the mobile phone signaling data, it can be known that the i-th mobile phone user is located at the j base station at time k, and the minimum outer rectangle generated by the Thiessen polygon constructed based on the j base station, and the random point is generated by the Monte Carlo simulation method. The probability that the i-th mobile phone user is located in the Thiessen polygon of base station j is set to 1, and the probability outside the Thiessen polygon and within the smallest enclosing rectangle is set to 0. Record the random point in the Thiessen polygon constructed by j base station as the specific base station position of the i-th mobile phone user at time k, and store the base station position corresponding to the i-th mobile phone user at time k through the mobile phone user ID . The above processing is performed on the base station where each user is located at each time, and the base station positions corresponding to each mobile phone user at each time are obtained.

According to the above steps, on the basis of the one-to-one corresponding base station positions of each mobile phone user at each time point, extract all the base station positions where each mobile phone user is at night (0-6 o'clock) time period, and calculate the location of each mobile phone user at 0-6 o'clock The proportion of staying at all base station locations, and calculate the total entropy value of each mobile phone user at night. If the total entropy value of the i-th mobile phone user is less than or equal to 1.5, then select the base station location where the i-th mobile phone user stays with the largest proportion of the night time period as the i-th mobile phone user's residence, and the i-th mobile phone user is outside the residence Other base stations are marked as the activity range of the i-th mobile phone user. In this embodiment, a total of 1,500,871 mobile phone users' residences are identified. The activity range and place of residence of each mobile phone user are associated with the mobile phone user ID.

Obtain the criminal information of the captured theft cases in ZG City from 2012 to June 2016, and use the online geocoding service interface to code the second place of crime and the second place of residence. A total of 282 criminals with more than 3 crimes in different areas of the city were screened out, and they were randomly divided into two groups, one group was a training sample of 133 people, and the other group was a test sample of 149 people, assuming that the test samples were the suspects to be predicted Suspects in the series.

Firstly, ZG city is spatially gridded, and ZG city is divided into 2562 square grids of 1609m×1609m, and each grid is numbered. For the suspect A in the series of cases to be predicted, grid numbering is carried out according to the first crime time and the first crime place. Extract the mobile phone users whose activity time and grid number of the activity range are the same as suspect A’s first committing time and the grid number of the first committing place in the processed data set of the activity range of all mobile phone users, and extract The place of residence of this part of the mobile phone users. Using the spatial kernel density analysis algorithm, generate the second kernel density map, the grid size is 100m×100m, and the search radius is 2000m.

The training sample has captured 133 criminals in a series of cases involving 467 cases. Calculate the Euclidean distance between the second place of crime and the second place of residence corresponding to the second place of crime in each case, and count the different distances The proportion of cases in the segment, with the segmented crime distance data as the horizontal axis and the case proportion as the vertical axis, constructs the crime distance distribution curve, and combines the negative exponential function to fit the crime distance distribution curve to obtain the crime distance attenuation function. As shown in Figure 4, the R square of the obtained fitting curve is 0.9935, indicating that the fitting degree is good.

Rasterize the research area, each grid size is 100m*100m. Calculate the distance from the center point of each grid to the first crime site, and substitute it into the crime distance attenuation function, calculate the probability value of each grid as the residence model of the suspect’s foothold to be predicted, and finally obtain the residence of the suspect to be predicted Earth models probability surface distributions.

For the first residence of all mobile phone users, use spatial kernel density analysis, set the search radius to 2000m, and set the grid size to 100m×100m to generate the first kernel density map of all mobile phone users. According to the framework of Bayesian theory, through grid calculation, the calculation method is as follows:

Among them, P(JTC|O) is the probability density of the suspect’s foothold to be predicted, P(JTC) is the probability surface distribution of the residence model of the suspect to be predicted, P(O|JTC) is the kernel density map of the second residence, and P (O) is the kernel density map of the first place of residence, and finally obtains the probability density map used to predict the whereabouts of the suspect to be predicted.

Obtain the test sample through this embodiment, that is, the probability density map of 149 suspects' footholds to be predicted, which is different from the traditional JTC model (expressed as JTC in Table 1) that only considers the distance decay curve, and the Bayesian model that only considers the OD matrix of captured criminals. The accuracy is compared between the Yessian JTC model (represented by CBJTC in Table 1) and the center point model (represented by Cmd in Table 1).

Accuracy evaluation includes the following three indicators:

Effectiveness: The ratio of the number of samples that can obtain the predicted results to the total test samples; the data that matches the predicted samples can be found from the training samples; the probability value of outputting the real suspect's foothold is not 0.

Search cost (accuracy): According to the probability density of the suspect’s whereabouts to be predicted, the police start searching from the highest point of the probability density until they find the area that needs to be searched for the real residence of the suspect to be predicted. The present invention uses the search cost ratio, that is, the ratio between the search area and the total area of ZG city.

Error distance (accuracy): the Euclidean distance between the highest point of the probability density of the suspect's foothold and the real place of residence of the suspect to be predicted. In this embodiment, the proportion of samples with an error distance<1609m (1 mile, the same grid size as ZG City) and an error distance<804.5m among the test samples is counted.

Table 1 The present invention compares with original method effect

It can be seen that the present invention realizes combining mobile phone signaling data with the traditional JTC model to predict the whereabouts of the suspect to be predicted by mining the spatial behavior patterns of criminals and mobile phone users, overcoming the fact that the traditional JTC model only considers a single distance factor, and further improves The precision and accuracy of predicting the whereabouts of the suspect to be predicted are improved. In the embodiment, compared with the traditional JTC model, this embodiment can reduce the search cost in the police investigation process by 56.74%, and the proportion of cases whose error distance is less than 1609 meters increases from 26.85% to 39.42%.

The invention can avoid the sparse problem of the OD matrix of the traditional Bayesian JTC model by fusing massive mobile phone signaling data and historical crime data. In the embodiment, the prediction accuracy of the traditional Bayesian JTC model is only 58.39%, while the prediction accuracy of this embodiment is 91.94%. The invention can effectively reduce the technical problem that a matching suspect cannot be found according to the first crime location and time of the serial cases of the suspect to be predicted, which leads to prediction failure. The invention can effectively improve the prediction precision and accuracy, so that the police can accurately predict the whereabouts of suspects.

The above description is a preferred embodiment of the present invention, it should be pointed out that for those skilled in the art, without departing from the principle of the present invention, some improvements and modifications can also be made, and these improvements and modifications are also considered Be the protection scope of the present invention.

Claims (7)

1. a kind of series case suspect's foothold prediction technique, it is characterised in that:

Obtain the first residence information and scope of activities information of all mobile phone users, wherein first residence information is used In the first residence cuclear density figure for generating all mobile phone users；

According to the scope of activities information, according to when and where, the first crime with the serial case of suspect to be predicted is extracted The mobile phone user that ground information matches, and according to the corresponding first residence information of the mobile phone user of extraction, in conjunction with preset sky Between cuclear density parser, obtain the second residence cuclear density figure of the suspect to be predicted；

According to the second of the preset serial case for having captured criminal the crime ground information and the second residence information, in conjunction with preset Negative exponential function obtains crime distance come the distance profile of with being fitted second crime the-the second residence of information information Attenuation function；

The inhabitation of suspect to be predicted is obtained in conjunction with the crime range-attenuation function algorithm according to first crime ground information The distribution of ground model probability surface；

According to the first residence cuclear density figure, the second residence cuclear density figure and residence model probability surface Distribution, in conjunction with preset bayesian algorithm, obtains the probability density figure for predicting suspect's foothold to be predicted.

2. series case suspect's foothold prediction technique as described in claim 1, which is characterized in that all mobile phones of the acquisition The the first residence information and scope of activities information of user, specifically:

First residence information includes several residences, the scope of activities information include several scopes of activities and with The scope of activities distinguishes the one-to-one activity time；

According to the mobile phone signaling data of all mobile phone users, extract, polymerization obtains whole base station point datas；

According to whole base stations point data, base station Thiessen polygon is constructed, and combine preset Monte Carlo simulation algorithm, obtained It obtains each mobile phone user and distinguishes one-to-one base station location at each moment；

Each base station location is calculated respectively using information Information Entropy, obtain all mobile phone users first occupies Residence information and scope of activities information.

3. series case suspect's foothold prediction technique as claimed in claim 2, which is characterized in that described to utilize information entropy Method respectively calculates each base station location, specifically:

Base station location corresponding to each mobile phone user's evening hours section is extracted, is calculated in each mobile phone user of evening hours section all The stop ratio of base station location；

According to the corresponding stop ratio of the entirety mobile phone user, night total entropy of each mobile phone user is calculated separately；

When the night of i-th of mobile phone user total entropy is less than preset threshold value, by the corresponding stop of i-th of mobile phone user The maximum base station location of ratio is labeled as the residence of i-th of mobile phone user, and i-th of mobile phone user is in residence Other base station locations in addition are labeled as the scope of activities of i-th of mobile phone user.

4. series case suspect's foothold prediction technique as described in claim 1, which is characterized in that the space cuclear density point Algorithm is analysed using four kernel functions.

5. series case suspect's foothold prediction technique as described in claim 1, which is characterized in that it is described according to it is preset The the second crime ground information and the second residence information for capturing the serial case of criminal, are fitted in conjunction with preset negative exponential function The distance profile of second crime ground the-the second residence of information information, obtains crime range-attenuation function, specifically:

Second crime ground information includes several the second crime ground, and second residence information includes to make with described second Distinguish one-to-one second residence to case；

It calculates separately each described second to commit a crime corresponding the distance between second residence, obtains crime distance Data；

Using the crime range data as horizontal axis, case ratio is the longitudinal axis, obtains crime distance profile；

It is fitted the crime distance profile using negative exponential function, obtains crime range-attenuation function f (d_ij):

d_ijPoint to the Euclidean distance of settlement, a_k、b_k、c_kParameter is constant value, need by the practical crime for having captured criminal away from It is obtained from distributed data training.

6. series case suspect's foothold prediction technique as described in claim 1, which is characterized in that described according to described first Crime ground information obtains the residence model probability surface point of suspect to be predicted in conjunction with the crime range-attenuation function algorithm Cloth, specifically:

First crime ground information includes several the first crime ground and respectively one-to-one the with first crime One crime time；

The calculation method for obtaining the residence model probability surface distribution of suspect to be predicted is as follows:

N*N meter of grid is generated in estimation range, calculates separately the distance that each grid central point commits a crime ground to described first x_ij, and by the x_ijIt is updated in the range-attenuation function, calculates inhabitation of each grid as suspect's foothold to be predicted Ground model probability value finally obtains the residence model probability surface distribution of suspect to be predicted.

7. series case suspect's foothold prediction technique as described in claim 1, which is characterized in that described according to described first Residence cuclear density figure, the second residence cuclear density figure and the distribution of residence model probability surface, in conjunction with preset Bayesian algorithm obtains the probability density figure for predicting suspect's foothold to be predicted, and calculation method is as follows:

Wherein, P (JTC | O) is the probability density of suspect's foothold to be predicted, and P (JTC) is the residence of the suspect to be predicted The distribution of residence model probability surface, and P (O | JTC) it is the second residence cuclear density figure, P (O) is the first inhabitation earth's core Density map.