A detection method and system based on multi-dimensional data fusion

Through the multi-dimensional data fusion method, mutation point detection and neural network technology are used to solve the problem of space-time relationship and dimensional disasters in massive information processing during investigation, accurate analysis and prediction of the whereabouts data of the target group are achieved, and investigation efficiency and accuracy are improved.

CN118797558BActive Publication Date: 2024-11-15HUNAN POLICE ACAD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411266602.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-11
Publication Date
2024-11-15
Estimated Expiration
2044-09-11

AI Technical Summary

Technical Problem

In the investigation work, it is difficult for the existing technology to effectively process and integrate massive diverse information, resulting in tracking errors, waste of resources and inaccurate investigation results, especially due to the lack of space-time relationships and dimensional disasters on the whereabouts of the target group.

Method used

Using a multi-dimensional data fusion method, through mutation point detection, correlation index and stationary factor calculation, combined with recurrent neural network and deep neural network, the whereabouts correlation coefficient and information contribution degree are obtained, and data fusion and prediction are carried out.

Benefits of technology

Accurate analysis and prediction of the whereabouts data of the target group is achieved, reducing the risk of dimensional disasters and improving the efficiency and accuracy of investigation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118797558B_ABST
    Figure CN118797558B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of data processing technology, and in particular to a detection method and system based on multidimensional data fusion, the method comprising: obtaining all whereabouts data of target persons in a target group; obtaining mutation points of each whereabouts data; analyzing the stability and correlation of each whereabouts data based on the mutation points to construct a whereabouts correlation coefficient; setting a reconnaissance period to construct a whereabouts correlation matrix; obtaining a whereabouts information contribution factor according to the correlation between different rows of the whereabouts correlation matrix; constructing the whereabouts information contribution of the target person; selecting the whereabouts data of the most representative target person in the target group to perform data fusion to obtain a whereabouts fusion vector of the target person; and completing the whereabouts prediction of the target person according to the whereabouts fusion vector to improve the representativeness of the fusion data and the accuracy of the predicted itinerary information.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a detection method and system based on multi-dimensional data fusion. Background Art

[0002] With the continuous development of the Internet, material networks, and sensor technology, various types of data continue to emerge, the amount of data is exploding, and the sources of data are becoming more and more diverse. Investigation work is also gradually moving towards informatization and intelligence, but how to extract the required data from massive amounts of multi-faceted information and integrate it to achieve intelligence is a huge challenge in the current investigation field.

[0003] In the current investigation work, when it is necessary to grasp the whereabouts of the target group, if the various data information is not comprehensively considered, it may cause the investigators to track the wrong target, wasting time and related resources; at the same time, if the target person's whereabouts information is lacking in the weight analysis of the contribution of the target group's whereabouts, the investigation results may be inaccurate. The whereabouts data of each target person in the target group includes real-time data such as the latitude and longitude of the person, the distance of the person's activities, the person's movement speed, the person's altitude, and the flow of people around the person. At present, the processing of whereabouts data mainly includes three categories: statistical analysis, anomaly detection, and machine learning. Among them, the method based on statistical analysis cannot capture the spatiotemporal relationship of whereabouts data, resulting in certain information loss; the whereabouts data analysis that relies on anomaly detection algorithms lacks accurate prior parameter experience values; when directly using machine learning algorithms to model and analyze the whereabouts data of all target persons in the target group, it will consume a lot of time and resources, and even cause dimensional disasters due to the multidimensionality of the data, making it difficult to provide short-term and efficient investigation data. Summary of the invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a detection method and system based on multi-dimensional data fusion. The technical solutions adopted are as follows:

[0005] In a first aspect, an embodiment of the present invention provides a detection method based on multi-dimensional data fusion, the method comprising the following steps:

[0006] Obtain all whereabouts data of each target person in the target group;

[0007] Perform mutation point detection on each of the target person's whereabouts data, and obtain the whereabouts correlation index of each mutation point of the whereabouts data according to the local data distribution of each mutation point of the whereabouts data and the change characteristics of all mutation points at the corresponding time before each mutation point; obtain the whereabouts stability factor of each mutation point of the whereabouts data according to the change characteristics of each mutation point of the whereabouts data and the local data distribution of all mutation points; obtain the whereabouts correlation coefficient of each whereabouts data according to the whereabouts correlation index and whereabouts stability factor of all mutation points of each whereabouts data;

[0008] Set a reconnaissance cycle, and obtain the whereabouts information contribution factor of each target person's whereabouts data according to the correlation of the whereabouts correlation coefficients of all target person's whereabouts data in different reconnaissance cycles; obtain the whereabouts information contribution degree of each target person according to the whereabouts information contribution factor of each whereabouts data;

[0009] According to the contribution of each target person's whereabouts information, a recurrent neural network is used to obtain the target person's whereabouts fusion vector; based on the whereabouts fusion vector, a deep neural network is used to complete the target person's whereabouts prediction.

[0010] Preferably, the obtaining of the whereabouts correlation index of each mutation point of the whereabouts data is specifically:

[0011] According to the distribution of local data of each mutation point of the whereabouts data, the expected difference of whereabouts of each mutation point of the whereabouts data is obtained;

[0012] The time of each mutation point of the whereabouts data and the corresponding time of all previous mutation points are taken as the whereabouts sudden change time sequence of each mutation point of the whereabouts data; the expected whereabout difference of each mutation point of the whereabouts data and the corresponding expected whereabout difference of all previous mutation points are taken as the expected whereabout difference sequence of each mutation point of the whereabouts data;

[0013] The whereabouts correlation index of the jth mutation point in the target person's i-th whereabouts data is recorded as , its formula form is: ;in, The number of types of target person's whereabouts data collected; , are the sudden change time sequences corresponding to the jth mutation point in the target person's i-th whereabouts data and the target person's p-th whereabouts data respectively; , are the expected difference sequences of whereabouts corresponding to the jth mutation point in the target person's i-th whereabouts data and the target person's p-th whereabouts data respectively; is the Jaccard distance function.

[0014] Preferably, the obtaining of expected differences in whereabouts of each mutation point in whereabouts data includes:

[0015] The mean of the whereabouts data at the moment before and after each mutation point is taken as the expected whereabouts value of each mutation point; the absolute value of the difference between the data value of each mutation point and the expected whereabouts value is taken as the expected whereabouts difference.

[0016] Preferably, the step of obtaining the whereabouts stability factor of each mutation point of the whereabouts data includes:

[0017] The absolute value of the difference between the expected difference of each mutation point and the mean of the expected difference sequence element is taken as the average expected difference;

[0018] Calculate the ratio of the number of elements in the expected difference sequence of whereabouts corresponding to each mutation point to the total number of data between the start time of data collection and the time corresponding to each mutation point;

[0019] The sum of the absolute values ​​of the differences between the expected whereabouts differences of all mutation points of the whereabouts data and the average expected difference is calculated; and the product of the ratio and the sum is used as the whereabouts stability factor of each mutation point of the whereabouts data.

[0020] Preferably, the whereabouts correlation coefficient of each item of whereabouts data is specifically the sum of the ratios of the whereabouts correlation index of all mutation points in each item of whereabouts data from the start time of data collection to the current time to the whereabouts stability factor.

[0021] Preferably, the acquisition of the whereabouts information contribution factor of each target person's whereabouts data specifically includes:

[0022] The mean of the whereabout correlation of each whereabout data in the reconnaissance period is taken as the element value of the whereabout correlation matrix of all target persons;

[0023] The whereabouts information contribution factor of the i-th whereabouts data of the x-th target person is recorded as , its formula form is: ;in, is the sum of the cosine similarities between the whereabouts correlation coefficient sequences of the x-th target person in the i-th whereabouts data collection interval and those of other target persons; Y is the total number of detection cycles in the whereabouts data collection interval of the target person; is the whereabouts correlation of the ith item of whereabouts data of the xth target person in the yth detection cycle; is the mean of the whereabouts correlation of the ith item of whereabouts data of the xth target person in all investigation cycles; e is a natural constant.

[0024] Preferably, the whereabouts information contribution of each target person is specifically the average of the whereabouts information contribution factors of all whereabouts data.

[0025] Preferably, the method of using a recurrent neural network to obtain the target person's whereabouts fusion vector according to the contribution of each target person's whereabouts information is as follows:

[0026] The whereabouts data of the target person with the greatest contribution to the whereabouts information are used as the input of the recurrent neural network; the whereabouts fusion vector of the target person is used as the output of the recurrent neural network; wherein each element of the whereabouts fusion vector is specifically the product of the mean of each whereabouts data in the reconnaissance period and the mean of the whereabouts correlation coefficient of the corresponding whereabouts data.

[0027] Preferably, the method of using a deep neural network to complete the whereabouts prediction of the target person according to the whereabouts fusion vector includes the following specific steps:

[0028] The target person's whereabouts fusion vector is used as the input of the deep neural network, and the target person's whereabouts information at the next moment is used as the output of the deep neural network; wherein the target person's whereabouts information at the next moment includes the target person's longitude, latitude and altitude data.

[0029] In a second aspect, an embodiment of the present invention further provides a detection system based on multi-dimensional data fusion, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of any one of the above-mentioned methods when executing the computer program.

[0030] The present invention has at least the following beneficial effects:

[0031] The present invention first analyzes the relevance of target persons in a target group to various whereabouts, and on the basis of obtaining the whereabouts correlation coefficient of the target person to each whereabouts data, further analyzes the whereabouts correlation coefficient of all target persons in the target group to each whereabouts data to obtain the contribution weight of each target person's whereabouts information, selects the most representative whereabouts information of the target person in the target group for data fusion, and finally predicts the whereabouts data of the target person after fusion.

[0032] At the same time, the present invention analyzes the relevant whereabouts data of all target persons in the target group, and reduces the dimension of the whereabouts data of the target group as much as possible on the basis of ensuring comprehensive consideration of various whereabouts data of the target persons, which can better avoid the occurrence of dimensional disasters; moreover, by comprehensively considering the stability and correlation between various whereabouts data of the target persons, a more accurate data fusion vector can be obtained according to the recurrent neural network, and then more accurate whereabouts prediction information can be obtained, thereby realizing a detection method and system based on multi-dimensional data fusion. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0034] Figure 1 A flowchart of a detection method based on multi-dimensional data fusion provided by an embodiment of the present invention;

[0035] Figure 2 Flowchart of the investigation method based on multi-dimensional data fusion. DETAILED DESCRIPTION

[0036] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation, structure, features and effects of a detection method and system based on multidimensional data fusion proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0037] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0038] The following is a detailed description of a specific scheme of a detection method and system based on multi-dimensional data fusion provided by the present invention in conjunction with the accompanying drawings.

[0039] See also Figure 1 , which shows a flowchart of a detection method based on multidimensional data fusion provided by an embodiment of the present invention, the method comprising the following steps:

[0040] Step S001: Obtain all the whereabouts data of each target person in the target group.

[0041] Specifically, all relevant target persons (i.e., investigation targets) whose whereabouts the investigators need to determine are recorded as a target group. The investigators collect the whereabouts data of each target person within 10km in the target group through the data monitoring system. The whereabouts data includes the longitude and latitude of the person, the distance of the person's activities, the speed of the person's movement, the altitude of the person, the flow of people around the person, and other relevant real-time whereabouts data. Among them, the longitude and latitude data of the target person can be obtained through the GPS positioning module in the data monitoring system; the distance of the target person's activities and the speed of movement can be obtained by calculating the actual position and passing time of the monitoring probe that the target person passes through in the data monitoring system; the altitude of the target person can be obtained through the actual altitude position of the monitoring probe in the data monitoring system; the flow of people around the target person can be determined by the sum of the number of all people in each monitoring probe with a neighborhood radius of 500m centered on the target person. The collection time interval of the whereabouts information is set to 1min, and the total collection time is time T. The corresponding time-series-based whereabouts data sequence is constructed according to the whereabouts data of each target person every minute within the total data collection time. In order to prevent the information collection module from malfunctioning and the loss of data during the transmission process, the collected whereabouts data sequence of the target person needs to be preprocessed.

[0042] The preprocessing algorithm used in this embodiment is the linear interpolation filling method, because the linear interpolation filling method does not change the distribution of the original data, and can use the trend of the existing data to infer the missing values, making the filled data more representative, and then the dimensional differences between the target person's whereabouts data are eliminated by the Z-score standardization method. The implementer can also select other algorithms to process the missing values ​​of the target person's whereabouts data series according to the actual situation. So far, all the target person's whereabouts data series based on time series after preprocessing can be obtained.

[0043] Step S002: Perform mutation point detection on each item of whereabouts data of the target person, and obtain the whereabouts correlation index of each mutation point of the whereabouts data according to the local data distribution of each mutation point of the whereabouts data and the change characteristics of all mutation points at corresponding moments before each mutation point; obtain the whereabouts stability factor of each mutation point of the whereabouts data according to the change characteristics of each mutation point of the whereabouts data and the local data distribution of all mutation points; obtain the whereabouts correlation coefficient of each item of whereabouts data according to the whereabouts correlation index and whereabouts stability factor of all mutation points of each item of whereabouts data.

[0044] When investigators determine the whereabouts data of the target group, the longitude position, latitude position, movement distance, movement speed, altitude, and surrounding environment flow of each target person in the target group that changes over time will affect the whereabouts data of the target group. When analyzing the whereabouts data of a single target person, it is impossible to accurately predict the whereabouts information of the target group; when directly analyzing the multidimensional data of the target group, it may cause a dimensional disaster, which will make it difficult to provide accurate auxiliary functions for investigators to provide accurate target group whereabouts information. Therefore, it is necessary to perform a fusion analysis based on the whereabouts data of the key target persons in the target group, while reducing the data dimension, considering the data characteristics of the multidimensional data itself and between the multidimensional data, so as to obtain more representative fusion data. Now, taking a target person in the target group as an example, the following analysis is performed on its whereabouts data:

[0045] First, the mutation data points in the target person's time-series-based whereabouts data are obtained through the Pettitt mutation point detection algorithm. The input of the Pettitt mutation point detection algorithm is the target person's time-series-based whereabouts data sequence, and the output is the mutation data point position in the target person's whereabouts data sequence (i.e., the moment at which the mutation data point is located in the target person's whereabouts data sequence). Since the Pettitt mutation point detection algorithm is a well-known technology, it will not be described in detail in this embodiment. The mean value of the whereabouts data item at the previous moment and the next moment of each whereabouts data item in the target person's whereabouts data sequence is calculated and recorded as the expected whereabouts value, and then the absolute value of the difference between the whereabouts data value of each mutation data point (i.e., the corresponding value at the mutation data point) and the expected whereabouts value at the mutation data point is calculated and recorded as the expected whereabouts difference .

[0046] In practical terms, the target person's whereabouts data based on time series will only be affected by the whereabouts data before the current moment. Therefore, the following analysis is performed on the mutation data points in the target person's whereabouts data sequence. According to all the whereabouts sudden change moments before the moment of each mutation data point in the target person's whereabouts data sequence (i.e., the moment of the mutation data point obtained above in the corresponding whereabouts data sequence) and the whereabouts expected difference, the whereabouts sudden change moment sequence and the whereabouts expected difference sequence of the mutation point are constructed. For example, the whereabouts sudden change moment sequence and the whereabouts expected difference sequence at the jth mutation point in the target person's i-th whereabouts data are respectively: , , are the moments of the mutation points before the jth mutation point in the target person’s i-th whereabouts data; are the mutation points before the jth mutation point in the target person's ith whereabouts data (i.e. The expected difference value of the whereabouts corresponding to the time.

[0047] The Jaccard distance between the sudden change time sequence corresponding to the jth mutation point of the target person's i-th whereabouts data and the sudden change time sequence corresponding to the jth mutation point of the target person's p-th whereabouts data is calculated and recorded as ; Calculate the Jaccard distance between the expected difference sequence of whereabouts corresponding to the jth mutation point of the target person's whereabouts data item i and the expected difference sequence of whereabouts corresponding to the jth mutation point of the target person's whereabouts data item p: It should be noted that the Jaccard distance is a known technology and will not be described in detail in this embodiment.

[0048] Based on the above data, the whereabouts association index is constructed, and the expression is:

[0049] ;

[0050] In the formula, is the whereabouts correlation index of the jth mutation point in the target person's i-th whereabouts data; The number of types of target person's whereabouts data collected; are the sudden change time sequences corresponding to the jth mutation point in the target person's i-th whereabouts data and the target person's p-th whereabouts data respectively; are the expected difference sequences of whereabouts corresponding to the jth mutation point in the target person's i-th whereabouts data and the target person's p-th whereabouts data respectively; is the Jaccard distance function.

[0051] At the same time, the absolute value of the difference between the expected difference of each mutation point and the mean of all data items in the expected difference sequence is calculated to obtain the mean expected difference MWA. The sudden change time sequence corresponding to the jth mutation point in the target person's i-th item of whereabouts data and the number of data elements in the expected difference sequence are recorded as (Both are constructed based on mutation points, so the number of data in the two sequences is the same), the total number of data between the start time of data collection and the jth mutation point in the target person's i-th whereabouts data sequence is calculated as , based on the above data, the tracking stability factor is constructed, and the expression is:

[0052] ;

[0053] In the formula, is the whereabouts stability factor of the jth mutation point in the target person's i-th whereabouts data; is the number of data elements in the sudden change time sequence corresponding to the jth mutation point in the target person's i-th whereabouts data and the expected whereabouts difference sequence; is the total number of data from the start time of data collection to the corresponding time of the jth mutation point in the target person's i-th whereabouts data sequence; is the expected difference of the whereabouts of the j mutation points in the target person's ith whereabouts data; It is the mean of all the expected whereabouts differences in the expected whereabouts difference sequence corresponding to the jth mutation point in the i-th whereabouts data of the target person.

[0054] The whereabouts correlation coefficient is constructed by combining the whereabouts association index and the whereabouts stability factor. The expression is:

[0055] ;

[0056] In the formula, is the whereabout correlation coefficient of the target person's ith whereabouts data; is the total number of mutation points in the ith item of whereabouts data from the start of data collection to the current time; is the whereabouts stability factor of the jth mutation point in the target person's i-th whereabouts data; is the whereabouts correlation index of the jth mutation point in the i-th whereabouts data of the target person.

[0057] When the relationship between the target person's whereabouts and the i-th item of whereabouts data is greater, the target person is more affected by the i-th item of whereabouts data. The larger the value is, the greater the impact on the target person's whereabouts is, the lower the stability of the ith item of whereabouts data is, and the smaller the whereabouts stability factor of the mutation point in the ith item of whereabouts data is, that is, The smaller is, the smaller is the proportion of mutation points in the data sequence in the i-th item of whereabouts data, that is, The smaller the expected difference of each mutation point in the i-th item of whereabouts data is, the smaller the difference in the data mean of the corresponding expected difference sequence of whereabouts is, that is, Similarly, when the relationship between the target person's whereabouts and the i-th whereabouts data is greater, the greater the impact of the whereabouts data, that is, The larger is the value, the greater the whereabouts correlation index of the mutation point in the ith whereabouts data, that is, The larger the value is, the smaller the Jaccard distance between the sudden change time sequence corresponding to the jth mutation point in the whereabouts data of the i-th target person and the sudden change time sequence corresponding to the jth mutation point in the whereabouts data of other target persons is, and the smaller the Jaccard distance between the whereabouts expected difference sequence corresponding to the jth mutation point in the whereabouts data of the i-th target person and the whereabouts expected difference sequence corresponding to the jth mutation point in the whereabouts data of other target persons is, the smaller the Jaccard distance between the whereabouts expected difference sequence corresponding to the jth mutation point in the whereabouts data of other target persons is. The bigger.

[0058] At this point, the whereabouts correlation coefficients of all target persons in the target group for each type of whereabouts data can be obtained according to the above method. In the target group, if the difference in the whereabouts correlation coefficients of a target person and other target persons obtained by analyzing the same type of whereabouts data is smaller, and the whereabouts correlation coefficients of the target person obtained by analyzing the same type of whereabouts data within the data collection interval are more stable, when determining the whereabouts information of the target group, the whereabouts information contribution weight of the target person is greater, and by analyzing the whereabouts correlations between different target persons, it is helpful to obtain more accurate location information of the target group.

[0059] Step S003: Set a reconnaissance period, and obtain the whereabouts information contribution factor of each target person's whereabouts data according to the correlation of the whereabouts correlation coefficients of all target persons' whereabouts data in different reconnaissance periods; and obtain the whereabouts information contribution degree of each target person according to the whereabouts information contribution factor of each whereabouts data.

[0060] In this embodiment, each hour is set as a reconnaissance cycle. The implementer can set the reconnaissance cycle according to the actual situation. The whereabouts correlation coefficient obtained by analyzing the same whereabouts data of all target persons in the target group within the whereabouts data collection interval is used to construct a whereabouts correlation matrix. The specific construction method is as follows: the whereabouts correlation coefficient of the ith whereabouts data of the xth target person in the yth reconnaissance cycle is averaged to obtain the whereabouts correlation of the whereabouts data, and a data sequence is constructed according to the whereabouts correlation obtained by analyzing the same whereabouts data of each target person in different reconnaissance cycles. The whereabouts correlation sequence obtained by analyzing the same whereabouts data of each target person within the whereabouts data collection interval is used as each row of the whereabouts correlation matrix. For example, the whereabouts correlation matrix of the ith whereabouts data of all target persons is:

[0061] ;

[0062] In the formula, It represents the mean correlation between the whereabouts of the x-th target person and the ith item of whereabouts data in the y-th detection cycle.

[0063] Calculate the mean of the whereabouts correlation of the xth target person's ith whereabouts data collection interval for all investigation cycles and record it as That is, the mean of each data item in the xth row of the above matrix is ​​calculated, and at the same time, the cosine similarity between the xth target person's ith item whereabouts data collection interval and the whereabouts correlation sequence of other target persons is calculated, that is, the sum of the cosine similarities between the xth row sequence and each of the remaining rows in the above matrix is ​​calculated, which is recorded as , based on the above characteristics, construct the whereabouts information contribution factor of the i-th whereabouts data of the x-th target person in the target group , the expression is:

[0064] ;

[0065] Where, Y is the total number of detection cycles within the target person’s whereabouts data collection interval; is the whereabouts correlation of the ith item of whereabouts data of the xth target person in the yth detection cycle; e is a natural constant.

[0066] Furthermore, the contribution of the whereabouts information of the xth target person is constructed , its formula form is:

[0067] ;

[0068] Where: is the number of types of target person's whereabouts data collected (corresponding to the number of types of target person's whereabouts data collected in this embodiment when calculating the whereabouts correlation index). consistent).

[0069] It should be understood that when determining the whereabouts information of the target group, the greater the contribution weight of the whereabouts information of the xth target person, that is, The larger the value is, the greater the contribution factor of the whereabouts information of the i-th whereabouts data of the x-th target person is, that is, The larger is the value, the smaller is the difference in the whereabout correlation between the target person and other target persons in the collected data interval, that is, The larger the value is, the smaller the difference between the whereabouts correlation of the i-th item of whereabouts data of the x-th target person in each investigation period and the mean whereabouts correlation of all investigation periods in all data collection intervals is, that is, The smaller.

[0070] At this point, the contribution of the whereabouts information of each target person in the target group can be obtained, and the whereabouts data of the target person with the largest contribution of whereabouts information is selected through the RNN recurrent neural network for data fusion. The input of the RNN recurrent neural network is the whereabouts data of the target person based on the time series. The optimization algorithm uses stochastic gradient descent SGD, and the loss function is the mean absolute error MAE. The output of the RNN recurrent neural network is the whereabouts fusion vector of the target person based on the time series. Each element of the whereabouts fusion vector is specifically the product of the mean of each whereabouts data in the reconnaissance period and the mean of the whereabouts correlation coefficient of the corresponding whereabouts data. It should be noted that since the RNN recurrent neural network is an existing well-known technology, it will not be described in detail in this embodiment.

[0071] Step S004: using a recurrent neural network to obtain a target person's whereabouts fusion vector based on the contribution of each target person's whereabouts information; and using a deep neural network to complete the target person's whereabouts prediction based on the whereabouts fusion vector.

[0072] In the above steps, the time-series-based whereabouts fusion vector of the most representative target person in the target group has been obtained, and the whereabouts information of the target person at the next moment is predicted by the deep neural network DNN. The whereabouts fusion vector of the target person is used as the input of the deep neural network DNN, Adam is used as the optimization algorithm, and cross entropy is used as the loss function. The output of the deep neural network DNN is the prediction information, that is, the whereabouts information of the target person at the next moment, including the longitude, latitude and altitude data of the target person. It should be noted that the deep neural network DNN is an existing well-known technology and will not be repeated in this embodiment.

[0073] Subsequent investigators can make decisions on the target group as quickly as possible with the help of the target person's latitude, longitude and altitude information at the next moment.

[0074] Thus, a detection method based on multi-dimensional data fusion is realized. Figure 2 shown.

[0075] Based on the same inventive concept as the above method, an embodiment of the present invention also provides a detection system based on multidimensional data fusion, including a memory, a processor, and a computer program stored in the memory and running on the processor, and when the processor executes the computer program, the steps of any one of the above-mentioned detection methods based on multidimensional data fusion are implemented.

[0076] In summary, the embodiment of the present invention first analyzes the degree of relevance of the target person in the target group to various whereabouts, and on the basis of obtaining the whereabouts correlation coefficient of the target person to each whereabouts data, further analyzes the whereabouts correlation coefficient of all target persons in the target group to each whereabouts data to obtain the contribution weight of each target person's whereabouts information, selects the most representative whereabouts information of the target person in the target group for data fusion, and finally predicts the whereabouts data of the target person after fusion.

[0077] At the same time, this embodiment analyzes the relevant whereabouts data of all target persons in the target group, and reduces the dimension of the whereabouts data of the target group as much as possible on the basis of ensuring comprehensive consideration of various whereabouts data of the target persons, which can better avoid the occurrence of dimensional disasters; moreover, by comprehensively considering the stability and correlation between various whereabouts data of the target persons, a more accurate data fusion vector can be obtained according to the recurrent neural network, and then more accurate whereabouts prediction information can be obtained, thereby realizing a detection method and system based on multi-dimensional data fusion.

[0078] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The above is a description of a specific embodiment of this specification. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0079] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0080] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A detection method based on multi-dimensional data fusion, characterized in that: The method comprises the following steps: Obtain all whereabouts data of each target person in the target group; Perform mutation point detection on each of the target person's whereabouts data, and obtain the whereabouts correlation index of each mutation point of the whereabouts data according to the local data distribution of each mutation point of the whereabouts data and the change characteristics of all mutation points at the corresponding time before each mutation point; obtain the whereabouts stability factor of each mutation point of the whereabouts data according to the change characteristics of each mutation point of the whereabouts data and the local data distribution of all mutation points; obtain the whereabouts correlation coefficient of each whereabouts data according to the whereabouts correlation index and whereabouts stability factor of all mutation points of each whereabouts data; Set a reconnaissance cycle, and obtain the whereabouts information contribution factor of each target person's whereabouts data according to the correlation of the whereabouts correlation coefficients of all target person's whereabouts data in different reconnaissance cycles; obtain the whereabouts information contribution degree of each target person according to the whereabouts information contribution factor of each whereabouts data; According to the contribution of each target person's whereabouts information, a recurrent neural network is used to obtain the target person's whereabouts fusion vector; according to the whereabouts fusion vector, a deep neural network is used to complete the target person's whereabouts prediction; The step of obtaining the whereabouts correlation index of each mutation point of the whereabouts data is specifically: According to the distribution of local data of each mutation point of the whereabouts data, the expected difference of whereabouts of each mutation point of the whereabouts data is obtained; The time of each mutation point of the whereabouts data and the corresponding time of all previous mutation points are taken as the whereabouts sudden change time sequence of each mutation point of the whereabouts data; the expected whereabout difference of each mutation point of the whereabouts data and the corresponding expected whereabout difference of all previous mutation points are taken as the expected whereabout difference sequence of each mutation point of the whereabouts data; The whereabouts correlation index of the jth mutation point in the target person's i-th whereabouts data is recorded as , its formula form is: ;in, The number of types of target person's whereabouts data collected; , are the sudden change time sequences corresponding to the jth mutation point in the target person's i-th whereabouts data and the target person's p-th whereabouts data respectively; , are the expected difference sequences of whereabouts corresponding to the jth mutation point in the target person's i-th whereabouts data and the target person's p-th whereabouts data respectively; is the Jaccard distance function; The acquisition of the whereabout information contribution factors of each target person's whereabouts data specifically includes: The mean of the whereabout correlation of all target persons' whereabouts data within the reconnaissance period is taken as the element value of the whereabout correlation matrix of all target persons; The whereabouts information contribution factor of the i-th whereabouts data of the x-th target person is recorded as , its formula form is: ;in, is the sum of the cosine similarities between the whereabouts correlation coefficient sequences of the x-th target person in the i-th whereabouts data collection interval and those of other target persons; Y is the total number of detection cycles in the whereabouts data collection interval of the target person; is the whereabouts correlation of the ith item of whereabouts data of the xth target person in the yth detection cycle; is the mean of the whereabouts correlation of the ith item of whereabouts data of the xth target person in all investigation cycles; e is a natural constant.

2. A detection method based on multidimensional data fusion as claimed in claim 1, characterized in that: The method of obtaining the expected difference of whereabouts at each mutation point of the whereabouts data includes: The mean of the whereabouts data at the moment before and after each mutation point is taken as the expected whereabouts value of each mutation point; the absolute value of the difference between the data value of each mutation point and the expected whereabouts value is taken as the expected whereabouts difference.

3. The detection method based on multidimensional data fusion as claimed in claim 1, characterized in that: The step of obtaining the whereabouts stability factor of each mutation point of the whereabouts data includes: The absolute value of the difference between the expected difference of each mutation point and the mean of the expected difference sequence element is taken as the average expected difference; Calculate the ratio of the number of elements in the expected difference sequence of whereabouts corresponding to each mutation point to the total number of data between the start time of data collection and the time corresponding to each mutation point; The sum of the absolute values ​​of the differences between the expected whereabouts differences of all mutation points of the whereabouts data and the average expected difference is calculated; and the product of the ratio and the sum is used as the whereabouts stability factor of each mutation point of the whereabouts data.

4. The detection method based on multi-dimensional data fusion as claimed in claim 1, characterized in that: The whereabout correlation coefficient of each item of whereabout data is specifically the sum of the ratios of the whereabout correlation index of all mutation points in each item of whereabout data from the start time of data collection to the current time to the whereabout stability factor.

5. The detection method based on multi-dimensional data fusion as claimed in claim 1, characterized in that: The whereabouts information contribution of each target person is specifically the average of the whereabouts information contribution factors of all whereabouts data.

6. The detection method based on multi-dimensional data fusion as claimed in claim 1, characterized in that: The method of using a recurrent neural network to obtain the target person's whereabouts fusion vector according to the contribution of each target person's whereabouts information is as follows: The whereabouts data of the target person with the greatest contribution to the whereabouts information are used as the input of the recurrent neural network; the whereabouts fusion vector of the target person is used as the output of the recurrent neural network; wherein each element of the whereabouts fusion vector is specifically the product of the mean of each whereabouts data in the reconnaissance period and the mean of the whereabouts correlation coefficient of the corresponding whereabouts data.

7. The detection method based on multi-dimensional data fusion as claimed in claim 1, characterized in that: The method of using a deep neural network to predict the whereabouts of a target person based on the whereabouts fusion vector includes the following specific steps: The target person's whereabouts fusion vector is used as the input of the deep neural network, and the target person's whereabouts information at the next moment is used as the output of the deep neural network; wherein the target person's whereabouts information at the next moment includes the target person's longitude, latitude and altitude data.

8. A detection system based on multi-dimensional data fusion, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.