Training method and device of gathering place model of specific personnel, medium and equipment

Abstract

The invention discloses a training method and device for a gathering place model of specific personnel, a medium and equipment, and belongs to the technical field of machine learning. Obtaining financial behavior characteristics, traffic behavior characteristics and geographic area characteristics of a specific crowd; a first training set is generated according to the financial behavior characteristics and the traffic behavior characteristics, training samples comprise financial behavior characteristics and traffic behavior characteristics in a first time period and financial behavior characteristics and traffic behavior characteristics in a second time period, and the second time period is later than the first time period; training an RNN model according to the first training set; a second training set is generated according to the financial behavior characteristics, the traffic behavior characteristics and the geographic area characteristics, training samples comprise the financial behavior characteristics, the traffic behavior characteristics, the geographic area characteristics and annotation information, and the annotation information represents the probability that a specific person appears in a certain gathering place in a certain geographic area at a certain time; and training the PSO-ELM model according to the second training set so as to improve the prediction accuracy.

Description

Technical Field

[0001] The present application relates to the field of machine learning technology, and in particular to a training method, device, medium and equipment for a gathering place model of specific people. Background Art

[0002] The determination of gathering places of specific people, such as drug addicts, is of great significance to public security, public health, etc. Accurate prediction of these gathering places can help law enforcement agencies more effectively supervise and combat illegal and criminal activities, and also help to take targeted prevention and intervention measures to reduce social risks.

[0003] Traditional methods for training models of gathering places of specific people mainly rely on manual investigation, reporting clues and limited data analysis. These methods have problems such as low efficiency, low accuracy and limited coverage. For example, manual investigation requires a lot of manpower and time costs, and it is difficult to fully cover all possible areas; reporting clues are uncertain and untimely; limited data analysis may not be able to fully tap potential information and it is difficult to accurately predict the gathering places of specific people. Summary of the invention

[0004] The present application provides a method, device, medium and equipment for training a gathering place model of specific people, which is used to solve the problems of low prediction efficiency, low accuracy and limited coverage of gathering places. The technical solution is as follows:

[0005] According to a first aspect of the present application, a method for training a gathering place model of specific people is provided, the method comprising:

[0006] Obtain financial behavior characteristics, transportation behavior characteristics, and geographic area characteristics of specific groups of people;

[0007] Generate a first training set according to the financial behavior characteristics and the traffic behavior characteristics, wherein the training samples in the first training set include the financial behavior characteristics and the traffic behavior characteristics of each specific person in a first time period and in a second time period, wherein the second time period is later than the first time period;

[0008] Training a recurrent neural network (RNN) model according to the first training set;

[0009] generating a second training set according to the financial behavior characteristics, the traffic behavior characteristics, and the geographic area characteristics, wherein the training samples in the second training set include the financial behavior characteristics, the traffic behavior characteristics, the geographic area characteristics, and annotation information of each specific person, wherein the annotation information indicates the probability of the specific person appearing at a certain gathering place in a certain geographic area at a certain time;

[0010] The particle swarm optimization algorithm PSO-extreme learning machine ELM model is trained according to the second training set.

[0011] In a possible implementation, the obtaining of financial behavior characteristics, traffic behavior characteristics, and geographic area characteristics of a specific group of people includes:

[0012] Obtaining financial behavior data, traffic behavior data, and geographic area data of a specific group of people, wherein the financial behavior data includes at least one of consumption location data, consumption time data, consumption amount data, and abnormal consumption frequency data; the traffic behavior data includes frequency data of specific people taking public transportation in a certain functional area, traffic flow data in and out of a transportation station, data on the functional area to which the transportation station belongs, and data on the correlation between travel peaks and suspicious gathering places; the geographic area data includes at least one of geographic data of each geographic area, population flow data, and historical crime data within the geographic area;

[0013] converting the financial behavior data into financial behavior features;

[0014] converting the traffic behavior data into traffic behavior features;

[0015] The geographic area data is converted into geographic area features.

[0016] In a possible implementation, converting the financial behavior data into financial behavior features includes:

[0017] For the consumption place data, the code of the category to which each consumption place belongs is combined into a first financial behavior feature;

[0018] For the consumption time data, the code of the time period to which each consumption time belongs is combined into a second financial behavior feature;

[0019] For the consumption amount data, the code of the consumption level to which each consumption amount belongs is combined into a third financial behavior feature;

[0020] For the abnormal consumption frequency data, the abnormal consumption frequency is calculated according to the abnormal consumption times and the total consumption times of each specific person, and the abnormal consumption frequency is formed into a fourth financial behavior feature;

[0021] The first financial behavior feature, the second financial behavior feature, the third financial behavior feature and the fourth financial behavior feature form a financial behavior feature.

[0022] In a possible implementation, converting the traffic behavior data into traffic behavior features includes:

[0023] For the frequency data of the specific person taking public transportation in a certain functional area, the number of times the specific person takes public transportation at each starting transportation station is counted, and the probability of the specific person taking public transportation in each functional area is calculated according to the number and the functional area to which each starting transportation station belongs, and the probabilities of all functional areas are combined into a first transportation behavior feature;

[0024] For the passenger flow data entering and leaving the traffic station, the passenger flow of each traffic station in different time periods is combined into a second traffic behavior feature;

[0025] For the functional area data to which the traffic station belongs, the codes of the functional areas to which each traffic station belongs are combined into a third traffic behavior feature;

[0026] For the correlation data between the travel peak and the suspicious gathering area, determine the probability that the business peak period of the gathering place where the specific person is located matches the travel peak period of the surrounding traffic stations, and form the fourth traffic behavior feature with the probability;

[0027] The first traffic behavior feature, the second traffic behavior feature, the third traffic behavior feature and the fourth traffic behavior feature form a traffic behavior feature.

[0028] In a possible implementation, training the RNN model according to the first training set includes:

[0029] Using the RNN model to process the financial behavior characteristics and traffic behavior characteristics in the first time period, to obtain predicted financial behavior characteristics and traffic behavior characteristics in the second time period;

[0030] Calculating a first loss value based on the predicted financial behavior characteristics and traffic behavior characteristics of the second period and the actual financial behavior characteristics and traffic behavior characteristics of the second period;

[0031] Update the model parameters of the RNN model according to the first loss value.

[0032] In a possible implementation manner, training the PSO-ELM model according to the second training set includes:

[0033] The PSO-ELM model is used to process the financial behavior characteristics and traffic behavior characteristics of the specific group of people to obtain a predicted value of the probability of the specific person appearing at a certain gathering place in a certain geographical area at a certain time;

[0034] Calculate the second loss value according to the predicted value and the labeled information;

[0035] The model parameters of the PSO-ELM model are updated according to the second loss value.

[0036] In a possible implementation, the method further includes:

[0037] Obtaining historical financial behavior characteristics and historical traffic behavior characteristics of a specific person to be predicted;

[0038] Using the RNN model to process the historical financial behavior characteristics and the historical traffic behavior characteristics, and predicting the future financial behavior characteristics and the future traffic behavior characteristics of the specific person;

[0039] The PSO-ELM model is used to process the historical financial behavior characteristics, the historical traffic behavior characteristics, the future financial behavior characteristics and the future traffic behavior characteristics to obtain the probability that the specific person will appear in a certain gathering place in a certain geographical area in the future.

[0040] According to a second aspect of the present application, a training device for a gathering place model of specific people is provided, the device comprising:

[0041] An acquisition module is used to acquire financial behavior characteristics, traffic behavior characteristics, and geographic area characteristics of a specific group of people;

[0042] A first generating module is used to generate a first training set according to the financial behavior characteristics and the traffic behavior characteristics, wherein the training samples in the first training set include the financial behavior characteristics and the traffic behavior characteristics of each specific person in a first time period and the financial behavior characteristics and the traffic behavior characteristics in a second time period, wherein the second time period is later than the first time period;

[0043] A first training module, used for training a recurrent neural network (RNN) model according to the first training set;

[0044] a second generating module, configured to generate a second training set according to the financial behavior characteristics, the traffic behavior characteristics and the geographic area characteristics, wherein the training samples in the second training set include the financial behavior characteristics, the traffic behavior characteristics, the geographic area characteristics and annotation information of each specific person, wherein the annotation information indicates the probability of the specific person appearing at a certain gathering place in a certain geographic area at a certain time;

[0045] The second training module is used to train the particle swarm optimization algorithm PSO-extreme learning machine ELM model according to the second training set.

[0046] According to a third aspect of the present application, a computer-readable storage medium is provided, in which at least one instruction is stored. The at least one instruction is loaded and executed by a processor to implement the training method of the gathering place model of specific personnel as described above.

[0047] According to a fourth aspect of the present application, a computer device is provided, which includes a training device for the above-mentioned gathering place model of specific persons.

[0048] The beneficial effects of the technical solution provided by this application include at least:

[0049] The training samples in the first training set include the financial behavior characteristics and traffic behavior characteristics of each specific person in the first time period and the financial behavior characteristics and traffic behavior characteristics in the second time period, where the second time period is later than the first time period. After training the RNN model according to the first training set, the RNN model can predict the future financial behavior characteristics and future traffic behavior characteristics based on the historical financial behavior characteristics and historical traffic behavior characteristics of the specific person, thereby expanding the behavioral characteristics of the specific person, providing more analytical data for predicting the behavior of the specific person, and improving the accuracy of the prediction.

[0050] The PSO-ELM model uses the PSO algorithm to optimize the parameters in the ELM. Specifically, the connection weights between the input layer and the hidden layer in the ELM and the thresholds of the neurons in the hidden layer are regarded as the particle positions in the PSO algorithm. Through the iterative search process of the PSO algorithm, a set of optimal parameters is found to improve the performance of the ELM. This combination can overcome the performance instability problem caused by random initialization of the ELM and improve the generalization ability and prediction accuracy of the PSO-ELM model. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0052] Figure 1 It is a flow chart of a method for training a gathering place model of specific persons provided by one embodiment of the present application;

[0053] Figure 2 It is a flow chart of a method for training a gathering place model of specific persons provided by one embodiment of the present application;

[0054] Figure 3 It is a flow chart of a method for training a gathering place model of specific persons provided by one embodiment of the present application;

[0055] Figure 4 It is a structural block diagram of a training device for a gathering place model of specific persons provided by an embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0057] like Figure 1 As shown, it shows a method flow chart of a method for training a gathering place model of specific people provided by an embodiment of the present application, and the training method for the gathering place model of specific people can be applied to a computer device. The training method for the gathering place model of specific people may include:

[0058] Step 101, obtaining financial behavior characteristics, traffic behavior characteristics and geographic area characteristics of a specific group of people.

[0059] A specific population is a group of specific people.

[0060] Financial behavior characteristics are characteristics extracted from the financial behavior generated by specific individuals. The specific extraction method is described in the following.

[0061] Traffic behavior features are features extracted based on the traffic behavior generated by specific people. The specific extraction method is described in the following.

[0062] The geographic region feature is a feature extracted based on the information of each geographic region. The specific extraction method is described in detail below.

[0063] Step 102, generating a first training set based on the financial behavior characteristics and the traffic behavior characteristics, wherein the training samples in the first training set include the financial behavior characteristics and the traffic behavior characteristics of each specific person in a first time period and in a second time period, wherein the second time period is later than the first time period.

[0064] For all financial behavior characteristics and traffic behavior characteristics of each specific person, the financial behavior characteristics and traffic behavior characteristics in the first period and the financial behavior characteristics and traffic behavior characteristics in the second period can be selected according to the time of occurrence of the behavior to form a training sample. For example, the first period is January 1, 2024, and the second period is February 5, 2024.

[0065] Step 103: train the RNN model according to the first training set.

[0066] The RNN model is a recurrent neural network (RNN) model.

[0067] When using each training sample to train the RNN model, the financial behavior characteristics and traffic behavior characteristics in the first period are used as the input of the RNN model, and the financial behavior characteristics and traffic behavior characteristics in the second period are used as the input of the loss function in the RNN model, so as to adjust the model parameters of the RNN model according to the output results of the loss function.

[0068] Step 104, generating a second training set based on the financial behavior characteristics, traffic behavior characteristics and geographic area characteristics, wherein the training samples in the second training set include the financial behavior characteristics, traffic behavior characteristics, geographic area characteristics and annotation information of each specific person, wherein the annotation information indicates the probability of the specific person appearing at a certain gathering place in a certain geographic area at a certain time.

[0069] For example, if a specific person often takes transportation to a bar near a public transportation station late at night, and the amount of consumption each time is large, the probability value in the annotation information can be set to be large. For another example, if a specific person takes transportation to a restaurant near a public transportation station at noon, and the amount of consumption each time is small, the probability value in the annotation information can be set to be small.

[0070] Step 105: training the PSO-ELM model according to the second training set.

[0071] The PSO-ELM model is a model composed of the Particle Swarm Optimization (PSO) algorithm and the Extreme Learning Machine (ELM).

[0072] When using each training sample to train the PSO-ELM model, the financial behavior characteristics, traffic behavior characteristics, and geographic area characteristics are used as the input of the PSO-ELM model, and the annotation information is used as the input of the loss function in the PSO-ELM model, so as to adjust the model parameters of the PSO-ELM model according to the output results of the loss function.

[0073] To sum up, the training method of the gathering place model of specific persons provided in the embodiment of the present application, the training samples in the first training set include the financial behavior characteristics and traffic behavior characteristics of each specific person in the first time period, and the financial behavior characteristics and traffic behavior characteristics in the second time period, the second time period is later than the first time period. After training the RNN model according to the first training set, the RNN model can predict the future financial behavior characteristics and future traffic behavior characteristics based on the historical financial behavior characteristics and historical traffic behavior characteristics of the specific person, thereby expanding the behavioral characteristics of the specific person, providing more analysis data for predicting the behavior of the specific person, and improving the accuracy of the prediction.

[0074] The PSO-ELM model uses the PSO algorithm to optimize the parameters in the ELM. Specifically, the connection weights between the input layer and the hidden layer in the ELM and the thresholds of the neurons in the hidden layer are regarded as the particle positions in the PSO algorithm. Through the iterative search process of the PSO algorithm, a set of optimal parameters is found to improve the performance of the ELM. This combination can overcome the performance instability problem caused by random initialization of the ELM and improve the generalization ability and prediction accuracy of the PSO-ELM model.

[0075] like Figure 2 As shown, it shows a flow chart of a method for training a gathering place model of specific people provided by an embodiment of the present application, and the training method for the gathering place model of specific people can be applied to a computer device. The training method for the gathering place model of specific people may include:

[0076] Step 201, obtaining financial behavior data, traffic behavior data and geographic area data of a specific group of people, wherein the financial behavior data includes at least one of consumption location data, consumption time data, consumption amount data and abnormal consumption frequency data; the traffic behavior data includes frequency data of specific persons taking public transportation in a certain functional area, flow data of people entering and leaving a transportation station, functional area data of the transportation station, and correlation data between travel peaks and suspicious gathering places; the geographic area data includes at least one of geographic data of each geographic area, population flow data and historical crime data within the geographic area.

[0077] Functional areas refer to the division of geographical areas according to their functions, such as commercial areas, residential areas, entertainment areas, industrial areas, etc.

[0078] Step 202: convert the financial behavior data into financial behavior features.

[0079] Specifically, converting financial behavior data into financial behavior features may include:

[0080] (1) For consumption location data, the code of the category to which each consumption location belongs is combined into the first financial behavior feature.

[0081] The dimension of the first financial behavior feature is equal to the number of categories of consumption venues. For example, if consumption venues are divided into 3 categories, the dimension of the first financial behavior feature is equal to 3, that is, the first financial behavior feature contains 3 elements.

[0082] Assume that the categories of consumption venues include restaurants (recorded as the first element in the first financial behavior feature), entertainment venues (such as bars, KTVs, etc., recorded as the second element in the first financial behavior feature), and retail stores (recorded as the third element in the first financial behavior feature). For example, if the bank account consumption records in a geographic area show that the proportion of consumption in entertainment venues is high, the geographic area may be more likely to have drug addicts gathering.

[0083] Assume that the consumption places are coded into three categories, and the first financial behavior feature of a specific person is represented in the form of a vector [0,0,1], where the third digit is 1, indicating that the main consumption place of the specific person belongs to the retail store category.

[0084] (2) For consumption time data, the code of the time period to which each consumption time belongs is combined into the second financial behavior feature.

[0085] Consumption time data is used to record the time period when consumption is concentrated. The dimension of the second financial behavior feature is equal to the number of time periods. For example, if the time periods are divided into 3 categories, the dimension of the second financial behavior feature is equal to 3, that is, the second financial behavior feature contains 3 elements.

[0086] Assume that the time periods of consumption are divided into daytime (06:00-18:00, recorded as the first element in the second financial behavior characteristic), evening (18:00-24:00, recorded as the second element in the second financial behavior characteristic), and late night (00:00-06:00, recorded as the third element in the second financial behavior characteristic). For example, if there are frequent consumption records late at night in a geographic area, it may indicate that there are some potential illegal activities including drug use in the geographic area.

[0087] Assume that consumption time is divided into three time periods, and the second financial behavior characteristic of a specific person is represented in the form of a vector [0,1,0], where the second digit is 1, indicating that the consumption of the specific person is mainly concentrated in the evening.

[0088] (3) For consumption amount data, the code of the consumption level to which each consumption amount belongs is combined to form the third financial behavior feature.

[0089] The consumption amount data is used to count the consumption frequency of different consumption levels (amount ranges). The dimension of the third financial behavior feature is equal to the number of consumption levels. For example, if there are 3 consumption levels, the dimension of the third financial behavior feature is equal to 3, that is, the third financial behavior feature contains 3 elements.

[0090] Assume that the consumption levels include small consumption (0-100 yuan, recorded as the first element in the third financial behavior characteristic), medium consumption (100-500 yuan, recorded as the second element in the third financial behavior characteristic), and large consumption (more than 500 yuan, recorded as the third element in the third financial behavior characteristic). For example, if a geographical area has a large number of large consumption records, and the consumption places are some suspicious entertainment places, this may be related to the high consumption behavior of drug users.

[0091] Assume that the consumption level is divided into three levels, and the third financial behavior characteristic of a specific person is expressed in the form of a vector [0.3, 0.4, 0.3], which means that the frequencies of small, medium and large consumption of the specific person are 0.3, 0.4 and 0.3 respectively.

[0092] (4) For abnormal consumption frequency data, the abnormal consumption frequency is calculated based on the abnormal consumption times and total consumption times of each specific person, and the abnormal consumption frequency is composed of the fourth financial behavior feature.

[0093] Abnormal consumption frequency data is used to identify the proportion of abnormal consumption times that are inconsistent with the average consumption habits of individuals or geographic areas to the total consumption times. For example, if an account suddenly has frequent high-value consumption, and the consumption location is in an area with complex public security, this may be abnormal consumption related to drug users buying drugs.

[0094] The fourth financial behavior characteristic can be represented by a numerical value, such as 0.2, which means that the frequency of abnormal consumption is 20%.

[0095] (5) The first financial behavior characteristic, the second financial behavior characteristic, the third financial behavior characteristic and the fourth financial behavior characteristic are combined into a financial behavior characteristic.

[0096] The computer device can set a splicing order of the first financial behavior feature, the second financial behavior feature, the third financial behavior feature and the fourth financial behavior feature, and splice the first financial behavior feature, the second financial behavior feature, the third financial behavior feature and the fourth financial behavior feature into a financial behavior feature according to the splicing order.

[0097] Step 203: convert the traffic behavior data into traffic behavior features.

[0098] Specifically, converting traffic behavior data into traffic behavior features may include:

[0099] (1) For the frequency data of a specific person taking public transportation in a certain functional area, the number of times the specific person takes public transportation at each starting transportation station is counted, and the probability of the specific person taking public transportation in each functional area is calculated based on the number of times and the functional area to which each starting transportation station belongs. The probabilities of all functional areas are combined into the first transportation behavior feature.

[0100] Assume that the functional areas are divided into four categories, namely commercial area (recorded as the first element in the first traffic behavior characteristic), residential area (recorded as the second element in the first traffic behavior characteristic), entertainment area (recorded as the third element in the first traffic behavior characteristic), and industrial area (recorded as the fourth element in the first traffic behavior characteristic). Computer equipment can classify the starting transportation stations of public transportation such as subways and buses according to the types of functional areas. Assume that there are 10 different starting transportation stations, and count the frequency of occurrence of the functional area to which each starting transportation station belongs.

[0101] Assuming that the first traffic behavior feature of a specific person is expressed in the form of a vector [0.2, 0.3, 0.4, 0.1], it means that in the statistical riding records, the frequency of the starting traffic station being a commercial area is 0.2, the probability of being a residential area is 0.3, the probability of being an entertainment area is 0.4, and the probability of being an industrial area is 0.1.

[0102] (2) For the flow of people in and out of traffic stations, the flow of people at each traffic station in different time periods is combined into the second traffic behavior feature.

[0103] Computer equipment counts the number of people entering and leaving each transportation station (bus station, subway station) at different time periods. For example, if the number of people at a subway station suddenly increases late at night and there are many entertainment venues around the geographical area, this may increase the possibility of drug users gathering.

[0104] (3) For the functional area data of the traffic station, the codes of the functional area to which each traffic station belongs are combined to form the third traffic behavior feature.

[0105] Assume that the functional areas are divided into four categories, namely, commercial areas (recorded as the first element in the third traffic behavior characteristic), residential areas (recorded as the second element in the third traffic behavior characteristic), entertainment areas (recorded as the third element in the third traffic behavior characteristic), and industrial areas (recorded as the fourth element in the third traffic behavior characteristic). Computer equipment can classify transportation stations of public transportation such as subways and buses according to the types of functional areas. For example, drug users may be more likely to gather around transportation stations located in entertainment areas.

[0106] (4) For the correlation data between travel peaks and suspicious gathering areas, determine the probability that the peak business hours of the gathering places where specific people are located match the peak travel hours of surrounding transportation stations, and use the probabilities to form the fourth traffic behavior feature.

[0107] Record whether peak traffic hours match known suspected geographic areas where drug users may gather (determined based on historical data or other clues). For example, if traffic stops around a suspected geographic area also reach peak traffic during the peak hours of entertainment venues in the evening, this may increase the probability that drug users use transportation to gather in that geographic area.

[0108] (5) The first traffic behavior characteristic, the second traffic behavior characteristic, the third traffic behavior characteristic, and the fourth traffic behavior characteristic are combined into a traffic behavior characteristic.

[0109] The computer device can set a splicing order of the first traffic behavior feature, the second traffic behavior feature, the third traffic behavior feature and the fourth traffic behavior feature, and splice the first traffic behavior feature, the second traffic behavior feature, the third traffic behavior feature and the fourth traffic behavior feature into a traffic behavior feature according to the splicing order.

[0110] Step 204: convert the geographic area data into geographic area features.

[0111] (1) When the geographic area data include geographic information:

[0112] The first geographic area feature is generated according to the coordinate positions of each block in the geographic area, such as (10.2, 20.5). Because the gathering places of drug addicts may be related to the geographical location, some remote areas or areas with relatively weak public security management may be more likely to become drug abuse sites. These coordinates can help determine different geographic areas.

[0113] The second geographic area feature is generated based on the distance from the geographic area to the entertainment venues (bars, nightclubs, etc.), in meters, such as 500 meters, 1000 meters, etc. There may be a higher probability of drug addicts gathering around entertainment venues due to factors such as complex personnel flow.

[0114] The third geographic area feature is generated according to the distance of the geographic area from public transportation stations (bus stations, subway stations), such as 300 meters, 700 meters, etc. Places with convenient transportation may facilitate the entry and exit of drug users, thereby increasing the possibility of them becoming gathering places.

[0115] (2) When the geographic area data include population mobility data:

[0116] The fourth geographic area feature is generated according to the flow of people in each geographic area in different time periods (such as daytime and nighttime). The flow of people can be the number of people passing through per hour, such as 1000 people / hour, 500 people / hour, etc. Places with low flow of people and inconvenient supervision may be more likely to have drug addicts gathering.

[0117] The fifth geographical area feature is generated based on the proportion of foreign population in the geographical area. The proportion of foreign population can be expressed as a percentage, such as 20%, 30%, etc. The probability of drug addicts gathering in geographical areas with a large number of foreign populations may also be affected due to factors such as complex population structure.

[0118] (3) When the geographic area data include historical crime data:

[0119] The sixth geographic region feature is generated based on the number of drug crime cases in the geographic region in the past year, such as 0, 3, etc. Geographic regions with a high incidence of drug crimes in history are likely to be gathering places for drug users.

[0120] The seventh geographic region feature is generated based on the number of other criminal cases (such as theft, robbery, etc.) in the geographic region. Because criminal behaviors are often related, a geographic region with a high crime rate may also have a relatively high concentration of drug users, with the number of cases being 5, 10, etc.

[0121] The computer device stitches the first geographic area feature, the second geographic area feature, the third geographic area feature, the fourth geographic area feature, the fifth geographic area feature, the sixth geographic area feature and the seventh geographic area feature into a geographic area feature.

[0122] Step 205, generating a first training set based on the financial behavior characteristics and the traffic behavior characteristics, wherein the training samples in the first training set include the financial behavior characteristics and the traffic behavior characteristics of each specific person in a first time period and in a second time period, wherein the second time period is later than the first time period.

[0123] The generation of the first training set is described in step 102 and will not be repeated here.

[0124] Step 206: train the RNN model according to the first training set.

[0125] When using each training sample to train the RNN model, the financial behavior characteristics and traffic behavior characteristics in the first period are used as the input of the RNN model, and the financial behavior characteristics and traffic behavior characteristics in the second period are used as the input of the loss function in the RNN model, so as to adjust the model parameters of the RNN model according to the output results of the loss function.

[0126] Specifically, training the RNN model according to the first training set may include: using the RNN model to process the financial behavior characteristics and traffic behavior characteristics in the first time period to obtain predicted financial behavior characteristics and traffic behavior characteristics of the second time period; calculating a first loss value based on the predicted financial behavior characteristics and traffic behavior characteristics of the second time period and the actual financial behavior characteristics and traffic behavior characteristics of the second time period; and updating the model parameters of the RNN model according to the first loss value.

[0127] Step 207, generating a second training set based on the financial behavior characteristics, traffic behavior characteristics and geographic area characteristics, wherein the training samples in the second training set include the financial behavior characteristics, traffic behavior characteristics, geographic area characteristics and annotation information of each specific person, wherein the annotation information indicates the probability of the specific person appearing at a certain gathering place in a certain geographic area at a certain time.

[0128] The generation of the second training set is described in step 104 and will not be repeated here.

[0129] Step 208: training the PSO-ELM model according to the second training set.

[0130] When using each training sample to train the PSO-ELM model, the financial behavior characteristics, traffic behavior characteristics, and geographic area characteristics are used as the input of the PSO-ELM model, and the annotation information is used as the input of the loss function in the PSO-ELM model, so as to adjust the model parameters of the PSO-ELM model according to the output results of the loss function.

[0131] Specifically, training the PSO-ELM model according to the second training set may include: using the PSO-ELM model to process the financial behavior characteristics and traffic behavior characteristics of a specific group of people to obtain a predicted value of the probability of a specific person appearing at a gathering place in a certain geographical area at a certain time; calculating a second loss value according to the predicted value and the annotation information; and updating the model parameters of the PSO-ELM model according to the second loss value.

[0132] To sum up, the training method of the gathering place model of specific persons provided in the embodiment of the present application, the training samples in the first training set include the financial behavior characteristics and traffic behavior characteristics of each specific person in the first time period, and the financial behavior characteristics and traffic behavior characteristics in the second time period, the second time period is later than the first time period. After training the RNN model according to the first training set, the RNN model can predict the future financial behavior characteristics and future traffic behavior characteristics based on the historical financial behavior characteristics and historical traffic behavior characteristics of the specific person, thereby expanding the behavioral characteristics of the specific person, providing more analysis data for predicting the behavior of the specific person, and improving the accuracy of the prediction.

[0133] The PSO-ELM model uses the PSO algorithm to optimize the parameters in the ELM. Specifically, the connection weights between the input layer and the hidden layer in the ELM and the thresholds of the neurons in the hidden layer are regarded as the particle positions in the PSO algorithm. Through the iterative search process of the PSO algorithm, a set of optimal parameters is found to improve the performance of the ELM. This combination can overcome the performance instability problem caused by random initialization of the ELM and improve the generalization ability and prediction accuracy of the PSO-ELM model.

[0134] like Figure 3 As shown, it shows a flowchart of a method for using a gathering place model of specific people provided by an embodiment of the present application, and the method for using the gathering place model of specific people can be applied to a computer device. The method for using the gathering place model of specific people may include:

[0135] Step 301, obtaining historical financial behavior characteristics and historical traffic behavior characteristics of a specific person to be predicted.

[0136] The computer device can obtain the historical financial behavior and historical traffic behavior of a specific person, extract historical financial behavior features from the historical financial behavior, and extract historical traffic behavior data from the historical traffic behavior. The specific extraction method is described in steps 202 and 203, which will not be repeated here.

[0137] Step 302: Use the RNN model to process the historical financial behavior characteristics and the historical traffic behavior characteristics, and predict the future financial behavior characteristics and the future traffic behavior characteristics of the specific person.

[0138] The computer device inputs the historical financial behavior characteristics and the historical traffic behavior characteristics into the RNN model, and determines the output of the RNN model as the future financial behavior characteristics and the future traffic behavior characteristics.

[0139] Step 303, using the PSO-ELM model to process the historical financial behavior characteristics, the historical traffic behavior characteristics, the future financial behavior characteristics and the future traffic behavior characteristics, to obtain the probability that a specific person will appear in a certain gathering place in a certain geographical area in the future.

[0140] The computer device inputs historical financial behavior characteristics, historical traffic behavior characteristics, future financial behavior characteristics and future traffic behavior characteristics into the PSO-ELM model, and determines the output of the PSO-ELM model as the probability that a specific person will appear in a certain gathering place in a certain geographic area in the future.

[0141] like Figure 4 As shown, it shows a structural block diagram of a training device for a gathering place model of specific people provided by an embodiment of the present application, and the training device for the gathering place model of specific people can be applied to a computer device. The training device for the gathering place model of specific people may include:

[0142] An acquisition module 410 is used to acquire financial behavior characteristics, traffic behavior characteristics, and geographic area characteristics of a specific group of people;

[0143] A first generating module 420, configured to generate a first training set according to the financial behavior characteristics and the traffic behavior characteristics, wherein the training samples in the first training set include the financial behavior characteristics and the traffic behavior characteristics of each specific person in a first time period and in a second time period, wherein the second time period is later than the first time period;

[0144] A first training module 430, used for training a recurrent neural network RNN ​​model according to a first training set;

[0145] A second generation module 440 is used to generate a second training set according to the financial behavior characteristics, the traffic behavior characteristics and the geographic area characteristics, wherein the training samples in the second training set include the financial behavior characteristics, the traffic behavior characteristics, the geographic area characteristics and the annotation information of each specific person, wherein the annotation information indicates the probability of the specific person appearing at a certain gathering place in a certain geographic area at a certain time;

[0146] The second training module 450 is used to train the PSO-ELM model according to the second training set.

[0147] In an optional embodiment, the acquisition module 410 is further configured to:

[0148] Obtaining financial behavior data, traffic behavior data, and geographic area data of specific people. Financial behavior data includes at least one of consumption location data, consumption time data, consumption amount data, and abnormal consumption frequency data. Traffic behavior data includes frequency data of specific people taking public transportation in a certain functional area, traffic flow data in and out of traffic stations, data on the functional areas to which traffic stations belong, and data on the correlation between travel peaks and suspicious gathering places. Geographic area data includes at least one of geographic data of each geographic area, population flow data, and historical crime data within the geographic area.

[0149] Convert financial behavior data into financial behavior features;

[0150] Converting traffic behavior data into traffic behavior features;

[0151] Convert geographic data into geographic features.

[0152] In an optional embodiment, the acquisition module 410 is further configured to:

[0153] For consumption venue data, the code of the category to which each consumption venue belongs is combined into the first financial behavior feature;

[0154] For consumption time data, the code of the time period to which each consumption time belongs is used to form the second financial behavior feature;

[0155] For the consumption amount data, the code of the consumption level to which each consumption amount belongs is combined into the third financial behavior feature;

[0156] For abnormal consumption frequency data, the abnormal consumption frequency is calculated based on the abnormal consumption times and total consumption times of each specific person, and the abnormal consumption frequency is composed of the fourth financial behavior feature;

[0157] The first financial behavior characteristic, the second financial behavior characteristic, the third financial behavior characteristic and the fourth financial behavior characteristic form a financial behavior characteristic.

[0158] In an optional embodiment, the acquisition module 410 is further configured to:

[0159] For the frequency data of a specific person taking public transportation in a certain functional area, the number of times the specific person takes public transportation at each starting transportation station is counted, and the probability of the specific person taking public transportation in each functional area is calculated based on the number of times and the functional area to which each starting transportation station belongs, and the probabilities of all functional areas are combined into the first transportation behavior feature;

[0160] For the flow of people in and out of the traffic station, the flow of people at each traffic station in different time periods is combined into the second traffic behavior feature;

[0161] For the functional area data of the traffic station, the codes of the functional area to which each traffic station belongs are combined into a third traffic behavior feature;

[0162] For the correlation data between travel peaks and suspicious gathering areas, determine the probability that the business peak hours of the gathering places where specific people are located match the travel peak hours of surrounding transportation stations, and use the probabilities to form the fourth traffic behavior feature;

[0163] The first traffic behavior feature, the second traffic behavior feature, the third traffic behavior feature and the fourth traffic behavior feature form a traffic behavior feature.

[0164] In an optional embodiment, the first training module 430 is further configured to:

[0165] The financial behavior characteristics and traffic behavior characteristics in the first period are processed by using the RNN model to obtain the predicted financial behavior characteristics and traffic behavior characteristics in the second period;

[0166] Calculating a first loss value based on the predicted financial behavior characteristics and traffic behavior characteristics of the second period and the actual financial behavior characteristics and traffic behavior characteristics of the second period;

[0167] Update the model parameters of the RNN model according to the first loss value.

[0168] In an optional embodiment, the second training module 450 is further configured to:

[0169] The PSO-ELM model is used to process the financial behavior characteristics and traffic behavior characteristics of a specific group of people to obtain the predicted value of the probability of a specific person appearing at a certain gathering place in a certain geographical area at a certain time;

[0170] Calculate a second loss value based on the predicted value and the labeled information;

[0171] The model parameters of the PSO-ELM model are updated according to the second loss value.

[0172] In an optional embodiment, the acquisition module 410 is further used to acquire the historical financial behavior characteristics and historical traffic behavior characteristics of the specific person to be predicted;

[0173] A prediction module is used to process historical financial behavior characteristics and historical traffic behavior characteristics using an RNN model, and predict future financial behavior characteristics and future traffic behavior characteristics of a specific person;

[0174] The prediction module is also used to process historical financial behavior characteristics, historical traffic behavior characteristics, future financial behavior characteristics and future traffic behavior characteristics using the PSO-ELM model to obtain the probability that a specific person will appear in a certain gathering place in a certain geographical area in the future.

[0175] To sum up, in the training device for the gathering place model of specific persons provided in the embodiment of the present application, the training samples in the first training set include the financial behavior characteristics and traffic behavior characteristics of each specific person in the first time period, and the financial behavior characteristics and traffic behavior characteristics in the second time period, and the second time period is later than the first time period. After the RNN model is trained according to the first training set, the RNN model can predict the future financial behavior characteristics and future traffic behavior characteristics based on the historical financial behavior characteristics and historical traffic behavior characteristics of the specific person, thereby expanding the behavioral characteristics of the specific person, providing more analysis data for predicting the behavior of the specific person, and improving the accuracy of the prediction.

[0176] The PSO-ELM model uses the PSO algorithm to optimize the parameters in the ELM. Specifically, the connection weights between the input layer and the hidden layer in the ELM and the thresholds of the neurons in the hidden layer are regarded as the particle positions in the PSO algorithm. Through the iterative search process of the PSO algorithm, a set of optimal parameters is found to improve the performance of the ELM. This combination can overcome the performance instability problem caused by random initialization of the ELM and improve the generalization ability and prediction accuracy of the PSO-ELM model.

[0177] An embodiment of the present application provides a computer-readable storage medium, in which at least one instruction is stored. The at least one instruction is loaded and executed by a processor to implement the training method of the gathering place model of specific people as described above.

[0178] An embodiment of the present application provides a computer device, which includes a training device for a gathering place model of any specific person mentioned above.

[0179] It should be noted that: when the training device for the gathering place model of specific persons provided in the above-mentioned embodiment is training the gathering place model of specific persons, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the training device for the gathering place model of specific persons is divided into different functional modules to complete all or part of the functions described above. In addition, the training device for the gathering place model of specific persons provided in the above-mentioned embodiment and the training method embodiment for the gathering place model of specific persons belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0180] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0181] The above description is not intended to limit the embodiments of the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the protection scope of the embodiments of the present application.

Claims

1. A method for training a gathering place model of specific people, characterized in that: The method comprises: Obtain financial behavior characteristics, transportation behavior characteristics, and geographic area characteristics of specific groups of people; Generate a first training set according to the financial behavior characteristics and the traffic behavior characteristics, wherein the training samples in the first training set include the financial behavior characteristics and the traffic behavior characteristics of each specific person in a first time period and in a second time period, wherein the second time period is later than the first time period; Training a recurrent neural network (RNN) model according to the first training set; generating a second training set according to the financial behavior characteristics, the traffic behavior characteristics, and the geographic area characteristics, wherein the training samples in the second training set include the financial behavior characteristics, the traffic behavior characteristics, the geographic area characteristics, and annotation information of each specific person, wherein the annotation information indicates the probability of the specific person appearing at a certain gathering place in a certain geographic area at a certain time; The particle swarm optimization algorithm PSO-extreme learning machine ELM model is trained according to the second training set.

2. The method for training a gathering place model of specific people according to claim 1, characterized in that: The acquisition of financial behavior characteristics, traffic behavior characteristics and geographic area characteristics of a specific group of people includes: Obtaining financial behavior data, traffic behavior data, and geographic area data of a specific group of people, wherein the financial behavior data includes at least one of consumption location data, consumption time data, consumption amount data, and abnormal consumption frequency data; the traffic behavior data includes frequency data of specific people taking public transportation in a certain functional area, traffic flow data in and out of a transportation station, data on the functional area to which the transportation station belongs, and data on the correlation between travel peaks and suspicious gathering places; the geographic area data includes at least one of geographic data of each geographic area, population flow data, and historical crime data within the geographic area; converting the financial behavior data into financial behavior features; converting the traffic behavior data into traffic behavior features; The geographic area data is converted into geographic area features.

3. The method for training a gathering place model of specific people according to claim 2, characterized in that: The converting the financial behavior data into financial behavior features comprises: For the consumption place data, the code of the category to which each consumption place belongs is combined into a first financial behavior feature; For the consumption time data, the code of the time period to which each consumption time belongs is combined into a second financial behavior feature; For the consumption amount data, the code of the consumption level to which each consumption amount belongs is combined into a third financial behavior feature; For the abnormal consumption frequency data, the abnormal consumption frequency is calculated according to the abnormal consumption times and the total consumption times of each specific person, and the abnormal consumption frequency is formed into a fourth financial behavior feature; The first financial behavior feature, the second financial behavior feature, the third financial behavior feature and the fourth financial behavior feature form a financial behavior feature.

4. The method for training a gathering place model of specific people according to claim 2, characterized in that: The converting the traffic behavior data into traffic behavior features comprises: For the frequency data of the specific person taking public transportation in a certain functional area, the number of times the specific person takes public transportation at each starting transportation station is counted, and the probability of the specific person taking public transportation in each functional area is calculated according to the number and the functional area to which each starting transportation station belongs, and the probabilities of all functional areas are combined into a first transportation behavior feature; For the passenger flow data entering and leaving the traffic station, the passenger flow of each traffic station in different time periods is combined into a second traffic behavior feature; For the functional area data to which the traffic station belongs, the codes of the functional areas to which each traffic station belongs are combined into a third traffic behavior feature; For the correlation data between the travel peak and the suspicious gathering area, determine the probability that the business peak period of the gathering place where the specific person is located matches the travel peak period of the surrounding traffic stations, and form the fourth traffic behavior feature with the probability; The first traffic behavior feature, the second traffic behavior feature, the third traffic behavior feature and the fourth traffic behavior feature form a traffic behavior feature.

5. The method for training a gathering place model of specific people according to claim 1, characterized in that: The step of training the RNN model according to the first training set includes: Using the RNN model to process the financial behavior characteristics and traffic behavior characteristics in the first time period, to obtain predicted financial behavior characteristics and traffic behavior characteristics in the second time period; Calculating a first loss value based on the predicted financial behavior characteristics and traffic behavior characteristics of the second period and the actual financial behavior characteristics and traffic behavior characteristics of the second period; Update the model parameters of the RNN model according to the first loss value.

6. The method for training a gathering place model of specific people according to claim 1, characterized in that: The step of training the PSO-ELM model according to the second training set comprises: The PSO-ELM model is used to process the financial behavior characteristics and traffic behavior characteristics of the specific group of people to obtain a predicted value of the probability of the specific person appearing at a certain gathering place in a certain geographical area at a certain time; Calculate the second loss value according to the predicted value and the labeled information; The model parameters of the PSO-ELM model are updated according to the second loss value.

7. The method for training a gathering place model of specific people according to any one of claims 1 to 6, characterized in that: The method further comprises: Obtaining historical financial behavior characteristics and historical traffic behavior characteristics of a specific person to be predicted; Using the RNN model to process the historical financial behavior characteristics and the historical traffic behavior characteristics, and predicting the future financial behavior characteristics and the future traffic behavior characteristics of the specific person; The PSO-ELM model is used to process the historical financial behavior characteristics, the historical traffic behavior characteristics, the future financial behavior characteristics and the future traffic behavior characteristics to obtain the probability that the specific person will appear in a certain gathering place in a certain geographical area in the future.

8. A training device for a gathering place model of specific people, characterized in that: The device comprises: An acquisition module is used to acquire financial behavior characteristics, traffic behavior characteristics, and geographic area characteristics of a specific group of people; A first generating module is used to generate a first training set according to the financial behavior characteristics and the traffic behavior characteristics, wherein the training samples in the first training set include the financial behavior characteristics and the traffic behavior characteristics of each specific person in a first time period and the financial behavior characteristics and the traffic behavior characteristics in a second time period, wherein the second time period is later than the first time period; A first training module, used for training a recurrent neural network (RNN) model according to the first training set; a second generating module, configured to generate a second training set according to the financial behavior characteristics, the traffic behavior characteristics and the geographic area characteristics, wherein the training samples in the second training set include the financial behavior characteristics, the traffic behavior characteristics, the geographic area characteristics and annotation information of each specific person, wherein the annotation information indicates the probability of the specific person appearing at a certain gathering place in a certain geographic area at a certain time; The second training module is used to train the particle swarm optimization algorithm PSO-extreme learning machine ELM model according to the second training set.

9. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the training method for the gathering place model of specific personnel as described in any one of claims 1 to 7.

10. A computer device, characterized in that: The computer device includes: a training device for the gathering place model of specific people as described in claim 8.

Images