Data mining method based on multi-task data mining results and target association

Abstract

The present invention relates to a data mining method for associating data mining results with targets based on multiple tasks, including: obtaining a target library and a single data mining result under a single task; looping through the targets in the target library, judging the feature types of PRI in the complete PRI feature sequence of the data mining result and the complete PRI feature sequence of the current target, if the feature types of the PRIs of the two are both fixed types or both group-variable types, first calling a fixed / group-variable_target library comparison method to compare the relevance of the data mining result with the current target; if the feature types of the PRIs of the two are both staggered types, calling a staggered_target library comparison method to compare the relevance of the data mining result with the current target, and completing data mining according to the relevance comparison result. The present invention improves the integrity and correctness of the data mining result by fusing the data mining results and targets with certain relevance under multiple tasks.

Description

Technical Field

[0001] The present invention relates to the technical field of data mining, and in particular to a data mining method based on multi-task association between data mining results and targets. Background Art

[0002] At present, complex radar signals have the characteristics of diverse bandwidth changes, variable frequency domain parameters, and complex time domain waveforms. In order to complete the detection of complex radar signal waveforms, reconnaissance equipment is usually required to perform multiple reconnaissance missions in multiple time domains and multiple spatial domains to detect the radar multiple times, resulting in the complete radar signal being distributed in multiple reconnaissance data.

[0003] Currently, traditional sorting algorithms only process single-mission data and cannot use multiple reconnaissance data and data mining technology to completely restore radar signals, resulting in inaccurate and incomplete intelligence results. Summary of the invention

[0004] In view of the problem that traditional sorting algorithms that only use single-mission data produce inaccurate and incomplete intelligence results when restoring complex radar signals, the present invention provides a data mining method based on multi-task data mining results and target associations. The method aims to utilize multiple reconnaissance data to restore complex radar signals to the maximum extent through data mining technology and data fusion technology, so as to provide accurate electronic intelligence for combat.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A data mining method based on multi-task data mining results and target association includes the following steps:

[0007] Step 1: Obtain the target database and a single data mining result under a single task;

[0008] Step 2: Loop through the targets in the target library, determine the feature type of the PRI in the complete PRI feature sequence of the data mining result and the complete PRI feature sequence of the current target, if the feature types of the PRI of both are fixed type or group variable type at the same time, execute step 3; if the feature types of the PRI of both are staggered type at the same time, execute step 4:

[0009] Step 3: first call the fixed / group-variable_target library comparison method to compare the correlation between the data mining result and the current target. When the data mining result is correlated with the current target, merge the data mining result with the current target to complete data mining; when the data mining result is not completely correlated with the current target, call the multi-task-based fixed / group-variable_activity law comparison method to perform further correlation comparison between the data mining result and the current target, and complete data mining according to the correlation comparison result;

[0010] Step 4: Call the Candidate_Target Library comparison method to compare the correlation between the data mining results and the current target, and complete the data mining according to the correlation comparison results.

[0011] Compared with the prior art, the present invention has the following beneficial effects:

[0012] The data mining method proposed in the present invention first calculates the correlation between a single data mining result under a single task and a certain target in the target library through the PRI parameter features, and then continues to compare the correlation between the data mining results and targets with a certain correlation through the PRI parameter features under multiple tasks to determine whether to fuse the data, thereby improving the integrity and correctness of the data mining results. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A flowchart of a data mining method based on multi-task data mining results and target association according to the present invention;

[0014] Figure 2 Flow chart of the fixed / group variable-target library comparison method of the present invention;

[0015] Figure 3 It is a flow chart of the fixed / group-variable-activity rule comparison method based on multi-task in the present invention;

[0016] Figure 4 Flow chart of the uneven-target library comparison method in the present invention. DETAILED DESCRIPTION

[0017] The technical solution of the present invention will be described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0018] The present invention provides a comprehensive mining method for data mining results and target association based on multiple tasks. The method improves the integrity and correctness of data mining results by calculating the association between a single data mining result under a single task and other targets under multiple tasks.

[0019] like Figure 1As shown, the present invention provides a data mining method based on multi-task data mining results and target association, comprising the following steps:

[0020] Step 1: Get the target database and the single data mining results under a single task, where the PRI feature parameters of each target in the target database are available, including the PRI feature type and the complete PRI feature sequence {PRI 1 , PRI 2 , ……PRI m＇}; The data mining result is that in a single task, the screening range is first determined according to the carrier frequency characteristic value RF and the pulse width characteristic value PW of the radiation source, and the original full pulse data is screened according to the screening range to obtain the screened full pulse data, and then the PRI feature type of the radiation source is used. According to different PRI feature types, different data mining methods are used to perform sequence search and data mining on the screened full pulse data according to the PRI feature sequence of the radiation source, and finally the data mining result is obtained, including the PRI feature type of the radiation source and the complete PRI feature sequence {PRI 1 , PRI 2 , ……PRI n＇};

[0021] Step 2: Loop through the targets in the target library, determine the feature type of PRI in the complete PRI feature sequence of the data mining result and the complete PRI feature sequence of the current target, if the feature types of the PRI of both are fixed type or group variable type, execute step 2; if the feature types of the PRI of both are staggered type, execute step 3:

[0022] Step 3: First, call the fixed / group-variable_target library comparison method to compare the correlation between the data mining results and the current target. When the data mining results are correlated with the current target, merge the data mining results with the current target to complete the data mining. When the data mining results are not completely correlated with the current target, call the multi-task-based fixed / group-variable_activity law comparison method to further compare the correlation between the data mining results and the current target, and complete the data mining according to the correlation comparison results.

[0023] Step 4: Call the Candidate_Target Library comparison method to compare the correlation between the data mining results and the current target, and complete the data mining based on the correlation comparison results.

[0024] Fixed / variable_target library comparison method

[0025] The complete characteristics of the PRI sequence of the group variable type are as follows: 1 , PRI 2 , ……PRI k＇} Pick any PRI from the seti＇ value, so that it appears O' times in a row, and then randomly select a PRI j＇ value (j'≠i'), so that it appears P' times in a row, the number of O' and P' is greater than or equal to 1 and is not fixed.

[0026] Traverse the targets in the target library. If the feature type of the PRI in the complete PRI feature sequence of the current target and the feature type of the PRI in the complete PRI feature sequence of the data mining result are both fixed types or both group-variable types, start data comparison. First, call the fixed / group-variable_target library comparison method to compare the relevance between the data mining result and the current target. The following are the specific comparison steps:

[0027] Step 3.1: The complete PRI feature sequence of the data mining results {PRI 1 , PRI 2 , ……PRI n＇}, and the complete PRI feature sequence of the current target {PRI 1 , PRI 2 , ……PRI m＇} to determine whether it meets condition (i):

[0028] (i) The PRI feature value of the data mining result is equal to the PRI feature value of the current target within the tolerance range.

[0029] Through the above comparison, the number C' of PRI feature values ​​in the PRI feature sequence of the data mining result that are the same as the PRI feature value in the PRI feature sequence of the current target is counted.

[0030] Step 3.2: Compare the number C' with the total number N' of PRI feature values ​​of the data mining results and the total number M' of PRI feature values ​​of the current target. The comparison conditions and corresponding correlation comparison results are as follows:

[0031] (ii) When the number C' is equal to the total number N', it means that the PRI feature sequence of the current target includes the PRI feature sequence of the data mining result. At this time, the loop is jumped out, and the correlation comparison between the data mining result and the target in the target library is ended. The data mining result is not processed;

[0032] (iii) When the number C' is equal to the total number M', it means that the PRI feature sequence of the data result includes the PRI feature sequence of the current target, and the data mining result is completely associated with the current target. At this time, the loop is jumped out, the correlation comparison between the data mining result and the target in the target library is ended, and the data mining result is merged with the current target to complete the data mining;

[0033] (iv) When the number C' is equal to 0, it means that the data mining result is not associated with the current target. At this time, the correlation between the data mining result and other targets continues to be compared until the loop is jumped out or the loop traversal is completed;

[0034] (v) When the number C' is less than the total number N' or less than the total number M', and the number C' is not equal to 0, it means that the data mining results are not completely associated with the current target. At this time, the multi-task based fixed / group variable_activity rule comparison method is called to further compare the correlation between the data mining results and the current target.

[0035] In other cases, it is judged that the data mining results are not related to the current target.

[0036] A multi-task-based fixed / group-variable activity pattern comparison method

[0037] After the fixed / group-variable_target library comparison, if the data mining results are not completely associated with the current target, it is necessary to continue the data comparison based on the multi-task fixed / group-variable_activity law comparison method. The following are the specific steps of the multi-task fixed / group-variable_activity law comparison method:

[0038] Step 3.2.1: Perform sequence search on the data mining results and the current target respectively under the current task to obtain the corresponding search PRI feature sequence, where the PRI feature sequence searched by the data mining result is {PRI 1 , PRI 2 , ……PRI s＇}, the PRI signature sequence of the current target search is {PRI 1 , PRI 2 , ……PRI t＇}.

[0039] Step 3.2.2: Search PRI feature sequence {PRI 1 , PRI 2 , ……PRI s＇} and the complete PRI signature sequence of the current target {PRI 1 , PRI 2 , ……PRI m＇} to compare and count the search PRI feature sequences {PRI1, PRI2, ...PRI s＇} and the complete PRI signature sequence {PRI 1 , PRI 2 , ……PRI m＇}, and the complete PRI feature sequence of the data mining result {PRI 1 , PRI 2 , ……PRIn＇} and the search PRI feature sequence of the current target {PRI 1 , PRI 2 , ……PRI t＇} to compare and count the complete PRI feature sequence {PRI 1 , PRI 2 , ……PRI n＇} and search for PRI signature sequence {PRI 1 , PRI 2 , ……PRI t＇ The number of PRI eigenvalues ​​in srcUN_C' that are not equal.

[0040] Determine whether the data mining result and the search PRI feature sequence of the current target meet any one of the activity rule comparison conditions (vi) and (vii). If so, record the activity rule comparison conditions that the current task meets. The activity rule comparison conditions (vi) and (vii) are as follows:

[0041] (vi) the number S of PRI feature values ​​of the search PRI feature sequence of the data mining result is equal to the total number N' of PRI feature values ​​of the complete PRI feature sequence of the data mining result, and the number T' of PRI feature values ​​of the search PRI feature sequence of the current target is equal to the total number M' of PRI feature values ​​of the complete PRI feature sequence of the current target;

[0042] (vii)digUN_C' is not equal to 0, and srcUN_C' is not equal to 0.

[0043] If condition (ⅵ) is met, it means that under the current task, the data mining results and the current goal appear at the same time;

[0044] If condition (ⅶ) is met, it means that under the current task, the data mining results and the PRI feature values ​​of the current target do not overlap and appear at the same time.

[0045] Step 3.2.3: Loop through all tasks, repeat steps 3.2.1 and 3.2.2, and record the activity pattern comparison conditions that each task meets. When two or more tasks meet condition (ⅵ) or condition (ⅶ) at the same time, merge the data mining results with the current target to complete data mining. In other cases, it is judged that the data mining results are not related to the current target.

[0046] If there are two or more tasks that meet the condition (vi) of step 3.2.2, the final conclusion is the merger of this data mining result and the goal;

[0047] If there are two or more tasks that meet the condition (ⅶ) of step 3.2.2, the final conclusion is that the data mining result is merged with the goal;

[0048] Otherwise, in all other cases, it is judged that the data mining results are not related to the target.

[0049] Different target library comparison method

[0050] The complete characteristics of the PRI sequence of the staggered type are as follows: 1 , PRI 2 , ……PRI i PRI I＇} is an ordered set, extracted in sequence, and the extraction result can be simply expressed as PRI i＇ , PRI i＇+1 , PRI i＇+2 PRI I＇ , PRI 1 , PRI 2 PRI i＇-1 The above extraction results appear in a cycle.

[0051] If the PRI feature type in the complete PRI feature sequence of the data mining result and the PRI feature type in the complete PRI feature sequence of the current target are both of the staggered type, then data comparison begins. Because the staggered type PRI sequence is an ordered set, when comparing it with the PRI sequence of the current target, unlike the fixed or group-varied type, the order problem must be considered. The following are the specific steps of the staggered_target library comparison method:

[0052] Step 4.1: First traverse the PRI feature sequence of the data mining results {PRI 1 , PRI 2 , ……PRI i＇ PRI N＇}, take its PRI characteristic value PRI in turn i＇ , and then take the PRI feature sequence of the current target {PRI 1 , PRI 2 , ……PRI j＇ PRI M＇ PRI characteristic value PRI in} j＇ , PRI i＇ With PRI j＇ Compare the sizes one by one to determine whether they meet the condition (ⅷ):

[0053] (ⅷ)PRI i＇ With PRI j＇ Equal in size within tolerance.

[0054] If condition (ⅷ) is met, count C' is increased by 1, and then PRI is taken in turn. i＇ The next PRI characteristic value PRI i＇+1 and PRIj＇ The next PRI characteristic value PRI j＇+1 Compare the sizes to determine whether they meet condition (ⅷ). If they do, add 1 to the count C', and so on. Through the above comparisons, we can finally count the number C' of PRI feature values ​​in the PRI feature sequence of the data mining result that are the same as those in the PRI feature sequence of the current target.

[0055] Step 4.2: Compare the number C' with the total number N' of PRI feature values ​​of the data mining results and the total number M' of PRI feature values ​​of the current target. The comparison conditions and corresponding correlation comparison results are as follows:

[0056] (ix) When the number C' is equal to the total number N', it means that the PRI feature sequence of the current target includes the PRI feature sequence of the data mining result. At this time, the loop is jumped out, and the correlation comparison between the data mining result and the target in the target library is ended. The data mining result is not processed;

[0057] (ⅹ) When the number C' is equal to the total number M', it means that the PRI feature sequence of the data result includes the PRI feature sequence of the current target, and the data mining result is completely associated with the current target. At this time, the loop is jumped out, the correlation comparison between the data mining result and the target in the target library is ended, and the data mining result is merged with the current target to complete the data mining.

[0058] In other cases, it means that the data mining result is not related to the current target, so the data mining result continues to be compared with other targets.

[0059] The step 1 of the present invention of obtaining a single data mining result under a single task is specifically implemented by the following steps:

[0060] Step I: First, determine the screening range according to the carrier frequency characteristic value RF and the pulse width characteristic value PW of the radiation source, and screen the original full pulse data according to the screening range to obtain the screened full pulse data.

[0061] In this step, the maximum value RF of the carrier frequency characteristic value RF is calculated according to the carrier frequency characteristic value RF and the pulse width characteristic value PW of the radiation source. max and minimum RF min And the maximum value PW of the pulse width characteristic value PW max and minimum PW min , then with a value greater than RF min -300MHz and less than RF max +300MHz and greater than PW min -50μs and less than PW max +50μs range is used as the screening range, that is, (RF min -300MHz,RF max+300MHz)∩(PW min -50μs,PW max +50μs), the original full pulse data is filtered according to the filtering range, and the filtered full pulse data is the data used for sorting and mining by the radiation source.

[0062] Step II: Obtain the PRI (pulse repetition time interval) feature type of the radiation source. If the PRI feature type is a fixed type or a group-variable type, execute step III, and use the fixed / group-variable mining method to perform data mining on the screened full pulse data to obtain the mined sample data; if the PRI feature type is a staggered type, execute step IV, and use the staggered mining method to perform data mining on the screened full pulse data to obtain the mined sample data.

[0063] Finally, step V: determine the abnormal type of the carrier frequency characteristic value RF and the pulse width characteristic value PW in the mined sample data according to the set carrier frequency tolerance RF_Tol and pulse width tolerance PW_Tol, and store the abnormal values ​​of the carrier frequency characteristic value RF and the pulse width characteristic value PW. min -RF_Tol and less than RF max The carrier frequency characteristic value of +RF_Tol is judged as an RF abnormal value, and the abnormal type is set to "rf out of range". min -PW_Tol and less than PW max The pulse width characteristic value of +PW_Tol is determined to be a PW abnormal value, the abnormal type is set to "pw out of range", and the RF abnormal value and PW abnormal value are stored.

[0064] The specific steps of the fixed / group-variable mining method in step III and the uneven mining method in step IV are introduced below.

[0065] Fixed / group-variable mining methods:

[0066] The complete characteristics of the PRI sequence of the group variable type are: 1 , PRI 2 , ……PRI O} Pick any PRI from the set i value, so that it appears S times in a row, and then randomly select a PRI j value (j≠i) so that it appears T times in a row, the number of S and T is greater than or equal to 1 and is not fixed.

[0067] When the PRI sequence of group-variable type is incomplete, data mining is required. The following are the data mining steps when the PRI feature type is fixed type or group-variable type, such as Figure 2 As shown:

[0068] Step 31: Perform sequence search on the filtered full pulse data according to the PRI signature sequence of the radiation source, and record the number of occurrences of each PRI signature value in the PRI signature sequence in set Q. Set Q is a set of mappings, whose key is each PRI signature value and value is the number of occurrences of the PRI signature value. Set Q obtained according to the most original PRI signature sequence (i.e., the PRI signature sequence that has never been updated) is recorded as set Q old After searching, several segments of sample data are obtained, and the interval between each segment of sample data is the interval to be mined.

[0069] The time interval of the filtered full pulse data is T A ~T B , through sequence search, a total of N segments of sample data are obtained, and the time interval of the nth (n=1,2,…,N) segment of sample data is t n,A ~t n,B , where subscript n represents the nth segment of sample data, and subscripts A and B represent the start and end of the time interval respectively.

[0070] Step 32: Obtain the mining interval according to each segment of sample data. The following are the steps for obtaining the mining interval:

[0071] (1) Determine the starting time t of the first segment of sample data 1,A Is it consistent with the starting time T of the filtered full pulse data? A If they are not equal, then T A ~t 1,A Recorded as 1 excavation interval;

[0072] (2) Determine the end time t of the Nth segment of sample data N,B Is it consistent with the end time T of the filtered full pulse data? B If they are not equal, then t N,B ~T B Recorded as 1 excavation interval;

[0073] (3) Set t n,B ~t n+1,A (n=1,2,…,N-1) that is, t 1,B ~t 2,A ,…,t N-1,B ~t N,A Recorded as N-1 mining intervals;

[0074] The mining intervals obtained in steps (1), (2), and (3) are totaled to M, where the time interval of the mth (m = 1, 2, ..., M) mining interval is t m,a ~t m,b , where the subscript m represents the mth mining interval, and the subscripts a and b represent the start and end time intervals of the mining interval, respectively.

[0075] Next, the M mining intervals obtained in step 32 are circulated and prepared for mining.

[0076] Step 33: In the mth mining interval, take the arrival time as time t m,a The pulse is taken as the first pulse to be mined, with the arrival time as time t m,b The pulse of is taken as the last pulse to be mined, and data mining is performed on the mth mining interval through the following steps:

[0077] (4) Calculate the j-level (j=1, 2, 3) arrival time difference ΔTOA for the full pulse data in the m-th mining interval j :

[0078] Definition: When j = 1, the arrival time difference between two adjacent pulses is the first-order arrival time difference ΔTOA 1 , the expression is as follows:

[0079] ΔTOA 1 =TOA i+1 –TOA i ; i=1,2,…,N m -1 ①

[0080] In formula ①, i represents the i-th pulse in the interval, N m is the total number of pulses, where i is accumulated one by one.

[0081] When j=2, there is one pulse between the two pulses, and the arrival time difference between the two pulses is the secondary arrival time difference ΔTOA 2 , the expression is as follows:

[0082] ΔTOA 2 =TOA i+2 –TOA i ; i=1,3,…,N m -2 ②

[0083] In formula ②, i represents the i-th pulse in the interval, N m is the total number of pulses, where i is accumulated by 2.

[0084] When j=3, there are two pulses between them, and the arrival time difference between the two pulses is the third-level arrival time difference ΔTOA 3 , the expression is as follows:

[0085] ΔTOA 3 =TOA i+3 –TOA i ; i=1,4,…,N m -3 ③

[0086] In formula ③, i represents the i-th pulse in the interval, N m is the total number of pulses, where i is accumulated by 3.

[0087] The first-level time difference of arrival ΔTOA 1 , Secondary time difference of arrival ΔTOA 2 , Level 3 Time Difference of Arrival ΔTOA 3 PRI i,j , PRI i,j Represents the time interval between the i+j-th pulse and the i-th pulse.

[0088] (5) PRI i,j The values ​​with the same value within the tolerance range are grouped into the kth group (k = 1, 2, ..., K), where K is the total number of groups finally divided, and the time interval of the kth group is collectively referred to as PRI k , where each group is a mapping whose key is the same value of PRI k , value is PRI k Count the number of occurrences C k . Count the PRIs that appear most often k And recorded as the preliminary mining value PRI_S.

[0089] (6) Perform sequence search in this mining interval based on the preliminary mining value PRI_S. If one or more continuous pulse data segments can be obtained, called distance verification data, the preliminary mining value PRI_S is subjected to distance verification and size verification, where the arrival time of the first pulse data in the first distance verification data segment is recorded as t_S. a , the arrival time of the last pulse data in the last distance verification data is recorded as t_S b .

[0090] In the sample data, t m,a The time interval with the previous pulse is denoted as PRI_YA, and t m,b The time interval with the next pulse is recorded as PRI_YB. If m=1, then PRI_YA=0; if m=M, then PRI_YB=0.

[0091] The distance in distance verification refers to the arrival time difference between pulse data, and the size in size verification refers to the size of the preliminary mining value PRI_S. First, verify whether the distance of the preliminary mining value PRI_S meets conditions (i) and (ii):

[0092] (i)t m,a With t_S a The difference is not greater than 10 times the value of PRI_YA;

[0093] (ii)t m,b With t_Sb The difference is no greater than 10 times the PRI_YB value.

[0094] If the preliminary mining value PRI_S meets at least one of the conditions (i) and (ii), the distance verification is successful. At this time, if only condition (i) is met, PRI_YA is recorded as PRI_Y; if only condition (ii) is met, PRI_YB is recorded as PRI_Y; if both conditions (i) and (ii) are met, the eigenvalue closest to the preliminary mining value PRI_S between PRI_YA and PRI_YB is recorded as PRI_Y.

[0095] Next, verify whether the size of the preliminary mining value PRI_S meets conditions (iii) and (iv):

[0096] (iii) The preliminary mining value PRI_S is less than 5 times PRI_Y;

[0097] (iv) The preliminary mining value PRI_S is greater than 0.2 times PRI_Y.

[0098] If the preliminary mining value PRI_S meets both conditions (iii) and (iv) at the same time, the size verification is successful, otherwise the size verification fails. If both the distance verification and the size verification are successful, the preliminary mining value PRI_S is updated to the PRI feature sequence. If the verification is not successful, the preliminary mining value PRI_S is deleted from the mapping group, and steps (5) and (6) are repeated until the data mining of the mth mining interval is completed.

[0099] Step 34: Return to step 31, recalculate step 31 with the PRI signature sequence updated by step 33, that is, re-search the filtered full pulse data according to the updated PRI signature sequence, and record the number of occurrences of each PRI signature value in the updated PRI signature sequence in the set Q new and obtain a new mining interval.

[0100] Next, we use the number of occurrences of PRI eigenvalues ​​before and after mining to verify the impact, that is, we use the set Q new With the set Q old Perform impact verification, where impact verification refers to whether the updated PRI feature sequence will affect the search results of the original PRI feature value during sequence search.

[0101] The following is the use of set Q new With the set Q old Steps to do impact verification:

[0102] Verify the set Q new With the set Q old Whether the following conditions (v) and (vi) are met:

[0103] (v) In the set Q old With the set Q new When there are identical key values, the set Q new The corresponding value is greater than or equal to the set Q old The corresponding value;

[0104] (vi) In the set Q new The key value exists, and the set Q old When there is no identical key value, the set Q new The corresponding value is greater than 0.

[0105] If the set Q new With the set Q old If conditions (v) and (vi) are met at the same time, the impact verification is successful. At this time, the M mining intervals obtained by step 32 are updated with the new mining intervals, and step 33 is continued to perform data mining to finally obtain the mined sample data; if the impact verification fails, the mining interval is not updated, and the next mining interval, that is, the m+1th mining interval, is selected to continue mining, and finally the mining is terminated until no data is mined, and finally the mined sample data is obtained.

[0106] Difference mining method:

[0107] The complete characteristics of the PRI sequence of the staggered type are as follows: 1 , PRI 2 , ……PRI i PRI O} is an ordered set, extracted in sequence, and the extraction result can be simply expressed as PRI i , PRI i+1 , PRI i+2 PRI O , PRI 1 , PRI 2 PRI i-1 The above extraction results appear in a cycle.

[0108] When the PRI sequence of the staggered type is incomplete, data mining is required. The following are the steps for data mining when the PRI feature type is staggered. Figure 3 As shown:

[0109] Step 41: Perform sequence search on the filtered full pulse data according to the PRI feature sequence of the radiation source to obtain several segments of sample data. When a PRI feature value in the PRI feature sequence is always separated from the next PRI feature value by one or more pulse data, the index of the current PRI feature value in the PRI feature sequence is recorded as the mining index, for example, i Always with PRIi+1 Every other pulse data, PRI i The index i-1 in the feature sequence is recorded as the mining index. The mining index can be one or more, and these mining indexes are recorded as a mining sequence.

[0110] Step 42: Traverse the mining sequence, prepare for mining, select one of the mining indexes, and perform the following mining steps:

[0111] (1) According to the PRI feature value corresponding to the selected mining index, the PRI feature value is recorded as PRI_i, and the sample data is traversed to find the two adjacent pulses PDW1 and PDW2 that can calculate the feature value PRI_i. From formula ①, it can be seen that the time interval is the difference between the arrival time of the latter pulse and the arrival time of the previous pulse. At this time, the arrival time of the latter pulse PDW2 is recorded as the mining start time TOA x In the sample data, the arrival time of the next pulse data after the pulse PDW2 is recorded as the mining termination time TOA y .

[0112] Mining start time TOA x As a benchmark, mining the end time TOA y As the boundary, start calculating the time interval ΔTOA j , the expression is as follows:

[0113] ΔTOA j =TOA x+j –TOA x ; j = 1, 2, ..., yx ④

[0114] (2) The time interval ΔTOA j From small to large, at most 3 values ​​are selected as mining reserve values, and each mining reserve value must meet the following condition (ⅶ):

[0115] (vii) Time interval ΔTOA j The absolute value of the difference with the eigenvalue PRI_i is less than or equal to 0.3 times the eigenvalue PRI_i.

[0116] (3) The mining backup values ​​obtained after calculating step (1) and step (2) once are called a group of backup values. Then, the sample data is traversed to continue to find adjacent pulses that can calculate the characteristic value PRI_i, and then steps (1) and step (2) are calculated once to obtain the second group of backup values. And so on, repeat steps (1) and step (2) for multiple times until 10 groups of backup values ​​are obtained.

[0117] (4) The 10 groups of backup values ​​with the same value within the tolerance range are grouped as the kth group (k = 1, 2, ..., K), where K is the total number of groups finally divided, and the time interval of the kth group is collectively referred to as PRIk , where each group is a mapping whose key is the same value of PRI k , value is PRI k Count the number of occurrences C k . Count the PRIs that appear most often k It is recorded as the preliminary mining value PRI_S.

[0118] (5) Next, continuously verify the preliminary mining value PRI_S: continue to traverse the sample data, repeat step (1) once, and follow the time interval ΔTOA obtained in step (1) j From small to large, find the ΔTOA that is equal to the preliminary mining value PRI_S within the tolerance j If found, this is a successful verification. Repeat step (1) multiple times until the sample data traversal is completed. During this period, if the continuous verification is successful for more than 3 times (not including 3 times), the continuous verification of the preliminary mining value PRI_S is successful.

[0119] Step 43: If the continuous verification of the preliminary mining value PRI_S is successful, the preliminary mining value PRI_S is the mining value, and the preliminary mining value PRI_S is updated to the PRI feature sequence, and the pulse data of the preliminary mining value PRI_S is also updated to the sample data, and the index of the preliminary mining value PRI_S in the PRI feature sequence is set as the new mining index, and step 42 is calculated repeatedly to see whether new mining values ​​can be mined after PRI_S, until no new mining values ​​are mined, that is, mining is completed, and finally the mined sample data is obtained. If the continuous verification is not successful during the first mining, it means that there is no mining value under the mining index, then continue to traverse the mining sequence, select the next mining index, and repeat step 42 until no PRI feature value is mined, mining is completed, and finally the mined sample data, that is, the single data mining result under a single task, is obtained.

[0120] The following takes a certain data mining result and target library as an example to introduce the technical solution of the present invention in detail.

[0121] Table 1 shows the information of a target in the target library as the current target:

[0122] Table 1 Current target information table

[0123]

[0124] Table 2 shows the data mining results:

[0125] Table 2 Data mining result information table

[0126]

[0127] First, based on the data mining results and the information of the current target, it can be determined that the characteristic type of the PRI of the two is the same as the group-variable type. Next, the fixed / group-variable_target library comparison method and the multi-task-based fixed / group-variable_activity law comparison method are used to continue the calculation. The following are the specific implementation steps:

[0128] Step 1: Substitute {1500, 1000, 1300} and {1500, 2000, 2500} for comparison.

[0129] Step 2: The feature value that meets condition (i) is counted as 1500us, and the number C'=1. Compare C'=1 with the total number of PRI feature values ​​N'=3 of the data mining result and the total number of PRI feature values ​​M'=3 of the current target, and it meets condition (v). The data mining result is not completely associated with the current target. At this time, the fixed / group-variable_activity rule comparison method based on multi-task is called to continue further correlation comparison.

[0130] Step 3: Under the task named Task 1, perform sequence search on the data mining results and the current target respectively. The search PRI feature sequence of the data mining results is {1000}, and the search PRI feature sequence of the current target is {2000, 2500}.

[0131] Step 4: Statistically obtain digUN_C'=1, srcUN_C'=2. Determine that the data mining result and the search PRI feature sequence of the current target meet condition (vii) and record them.

[0132] Step 5: Perform comparisons under all tasks and record the activity rule comparison conditions that all tasks meet. The results are shown in Table 3.

[0133] Table 3. Comparison results of multi-task activity patterns

[0134]

[0135] As can be seen from Table 3, there are 3 tasks that meet condition (ⅶ), so the final conclusion is to merge the data mining result with the current goal to complete the data mining.

[0136] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0137] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A data mining method based on multi-task data mining results and target association, It is characterized in that The following steps are involved: Step 1: Obtain the target database and a single data mining result under a single task; Step 2: Loop through the targets in the target library, determine the feature type of the PRI in the complete PRI feature sequence of the data mining result and the complete PRI feature sequence of the current target, if the feature types of the PRI of both are fixed type or group variable type at the same time, execute step 3; if the feature types of the PRI of both are staggered type at the same time, execute step 4: Step 3: first call the fixed / group-variable_target library comparison method to compare the correlation between the data mining result and the current target. When the data mining result is correlated with the current target, merge the data mining result with the current target to complete data mining; when the data mining result is not completely correlated with the current target, call the multi-task-based fixed / group-variable_activity law comparison method to perform further correlation comparison between the data mining result and the current target, and complete data mining according to the correlation comparison result; The fixed / group-variable-target library comparison method comprises the following steps: Step 3.1: Count the number C' of PRI feature values ​​in the PRI feature sequence of the data mining result that are the same as the PRI feature value in the PRI feature sequence of the current target; Step 3.2: Compare the number C' with the total number N' of PRI feature values ​​of the data mining result and the total number M' of PRI feature values ​​of the current target, respectively. The comparison conditions and the corresponding correlation comparison results are as follows: (ii) When the number C' is equal to the total number N', it means that the PRI feature sequence of the current target includes the PRI feature sequence of the data mining result, and the loop is exited at this time, and the correlation comparison between the data mining result and the target in the target library is ended, and the data mining result is not processed; (iii) When the number C' is equal to the total number M', it means that the PRI feature sequence of the data mining result includes the PRI feature sequence of the current target, and the data mining result is completely associated with the current target. At this time, the loop is jumped out, the correlation comparison between the data mining result and the targets in the target library is ended, and the data mining result is merged with the current target to complete the data mining; (iv) When the number C' is equal to 0, it means that the data mining result is not associated with the current target. At this time, the correlation between the data mining result and other targets continues to be compared until the loop is exited or the loop traversal is completed; (v) When the number C' is less than the total number N' or less than the total number M', and the number C' is not equal to 0, it means that the data mining result is not completely associated with the current target. At this time, the fixed / group-variable activity rule comparison method based on multi-task is called to further compare the correlation between the data mining result and the current target; Step 4: Call the Candidate_Target Library comparison method to compare the correlation between the data mining results and the current target, and complete the data mining according to the correlation comparison results.

2. According to claim 1, a data mining method based on multi-task data mining results and target association, It is characterized in that The multi-task based fixed / group variable-activity rule comparison method comprises the following steps: Step 3.2.1: performing sequence search on the data mining result and the current target respectively under the current task to obtain the corresponding search PRI feature sequence; Step 3.2.2: Count the number of PRI feature values ​​in the search PRI feature sequence of the data mining result that are not equal to the PRI feature values ​​in the complete PRI feature sequence of the current target digUN_C', and count the number of PRI feature values ​​in the complete PRI feature sequence of the data mining result that are not equal to the PRI feature values ​​in the search PRI feature sequence of the current target srcUN_C'; Determine whether the data mining result and the search PRI feature sequence of the current target meet any one of the activity rule comparison conditions (vi) and (vii). If so, record the activity rule comparison condition that the current task meets, wherein the activity rule comparison conditions (vi) and (vii) are as follows: (vi) the number S' of search PRI feature sequence PRI feature values ​​of the data mining result is equal to the total number N' of complete PRI feature sequence PRI feature values ​​of the data mining result, and the number T' of search PRI feature sequence PRI feature values ​​of the current target is equal to the total number M' of complete PRI feature sequence PRI feature values ​​of the current target; (vii) digUN_C' is not equal to 0, and srcUN_C' is not equal to 0; Step 3.2.3: Loop through all tasks, and record the activity pattern comparison conditions that each task meets. When two or more tasks meet condition (ⅵ) or condition (ⅶ) at the same time, merge the data mining results with the current target to complete data mining. In other cases, it is judged that the data mining results are not related to the current target.

3. The data mining method based on multi-task data mining results and target association according to claim 1, It is characterized in that The uneven-target library comparison method comprises the following steps: Step 4.1: sequentially compare the PRI feature values ​​in the PRI feature sequence of the data mining result with the PRI feature values ​​in the PRI feature sequence of the current target, and count the number C' of identical PRI feature values; Step 4.2: Compare the number C' with the total number N' of PRI feature values ​​of the data mining result and the total number M' of PRI feature values ​​of the current target, respectively. The comparison conditions and the corresponding correlation comparison results are as follows: (ix) When the number C' is equal to the total number N', it means that the PRI feature sequence of the current target includes the PRI feature sequence of the data mining result, and the loop is exited at this time, and the correlation comparison between the data mining result and the target in the target library is ended, and the data mining result is not processed; (ⅹ) When the number C' is equal to the total number M', it means that the PRI feature sequence of the data mining result includes the PRI feature sequence of the current target, and the data mining result is completely associated with the current target. At this time, the loop is jumped out, the correlation comparison between the data mining result and the target in the target library is ended, and the data mining result is merged with the current target to complete the data mining.

Images