CN108415954A - The uncertain monochromatic mutually K-NN search processing method of one kind - Google Patents

Abstract

The present invention provides the uncertain monochromatic mutually K-NN search processing methods of one kind, and include step：Initialize most rickle H_rfn, enable set S_stemp、O_cand、O_pruDeng being empty；R trees are traversed, the arest neighbors Candidate Set nn of query object q is obtained_q, and most rickle H will be stored in by the node of beta pruning and object during this_rfnIt is interior；Successively by nn_qIn object o be inserted into H_rfnIn, while in set O_candIt is upper that object o is judged using GP rules, o is inserted into set O if o is eligible_pru, o is otherwise inserted into set O_cand；By H_rfnIn element be inserted into set S_stemp；In set O_pruIt is upper regular using GP, find out set O_candIn be not centainly q Reverse Nearest object, and make marks；To O_candIn each object o for not making marks, traversal set S_stempInquiry obtains the arest neighbors Candidate Set nn of o_o；Find out nn_oWith nn_qUnion obtain the probability calculation list of o；Calculate O_candIn the probability value of each object that does not make marks；The object that probability value is more than to threshold value returns as a result.

Description

The uncertain monochromatic mutually K-NN search processing method of one kind

Technical field

The invention belongs to space-time database technical fields, at the uncertain monochromatic mutually K-NN search of one kind Reason method.

Background technology

Space-time database is the important branch of database field, and spatial-temporal query is the important operation in space-time database, root It can be divided into according to steric requirements：Range query, K-NN search, reverse nearest neighbor queries etc..Mutual K-NN search is most The deformation and extension of NN Query, reverse nearest neighbor queries, result of calculation are the arest neighbors of set point, while also with set point For the data point of arest neighbors.Mutual K-NN search has emphatically in fields such as data mining, pattern-recognition and decision supports The application value wanted.

In monochromatic mutually K-NN search, inquires sender q and target object belongs to the same data set D.Therefore one A data object o is considered as the mutual arest neighbors for inquiring sender q, and if only if in data set between any object o ' and q Distance be more than o between q at a distance from, while o ' with o between at a distance from more than o between q at a distance from.I.e.Dist (q, o)≤dist (q, o ') ∧ dist (o, q)≤dist (o, o ') }, wherein dist () is the distance between two objects function.

In space-time database most common index structure be based on R trees or its derive from.R trees are one Kind height balanced tree, may be implemented complete dynamic index.The index structure that the present invention uses is R trees.

Due to the limitation of many factors such as location technology, the bandwidth of network, system storage, the data of space-time object can only be with Discrete way obtains.Further, since the characteristics of measurement error, concrete application (such as location privacy protection) factor makes space-time number It is often inconsistent with the real data of object according to the information stored in library.Locational uncertainty is the fields such as space-time database, LBS Unavoidable problem.It is general to indicate object at a time possible position range with an area of space, it is referred to as uncertain Domain indicates probability distribution of the object's position in uncertain domain with probability density function.

Doctoral thesis《Position does not know mobile object query processing key technology research》(Nanjing Aero-Space University, In October, 2013) in propose a kind of methods (being denoted as MN) based on more wheel nearest neighbor search technologies, asked although this can be solved Topic, but have the following defects：(1) to R trees repeatedly from root node repeated accesses, thus I/O costs are higher；(2) Candidate Set mistake Greatly, on the one hand cause unnecessary I/O operation, on the other hand increase the probability calculation amount in refinement step so that generation time Valence is excessively high.

In the patent of invention of Patent No. 201610118192.5《A kind of monochrome in Instable Space data is mutually nearest Adjacent inquiry processing method》In (being denoted as NR), propose：First, it avoids repeating to access R trees from root node using reuse technology, to Reduce the I/O operation in Reverse Nearest search and probability calculation list construction process；Secondly, pass through reverse nearest neighbor queries Friendship is asked with the result of K-NN search, smaller candidate target collection is obtained, had not only reduced probability calculation amount but also reduced I/O times Number.Relative to method MN, performance is improved.

But NR methods still have following improved space：(1) method NR needs to carry out reverse nearest neighbor queries, although Reuse technology is utilized, but I/O expenses are still higher, and a portion is avoidable；(2) it is looked into Reverse Nearest During inquiry, the exclusion of subsequent objects could not be used for by the object of GP rule prunings, this can also cause unnecessary I/O operation and Time cost.

Invention content

It is an object of the invention in view of the drawbacks of the prior art or problem, provide a kind of uncertain monochromatic mutual arest neighbors Inquiry processing method.

Technical scheme is as follows, the uncertain monochromatic mutually K-NN search processing method of one kind, including walks as follows Suddenly：

Step 1：Initialize most rickle H_rfn, enable set S_temp,O_cand,O_pruFor sky；

Step 2：Traverse R trees, update in the process and obtain all objects to the maximum distance of query object q minimum Value minf, and node and the object beta pruning that minf will be more than to the minimum distance of q, while being keyword by the minimum distance to q It will be preserved to most rickle H by the node of beta pruning and object_rfnIn, fail to be stored in set nn by the object of beta pruning_qIn；

Step 3：By nn_qIn each object o according to its to q minimum distance be keyword be inserted into H_rfnIn；Collecting simultaneously Close O_candIt is upper that object o is judged using GP rules, if o is eligible, o is inserted into set O_pru, otherwise o is inserted into Set O_cand；

Step 4：By H_rfnIn element be entirely insertable set S_temp。

Step 5：In set O_pruIt is upper regular using GP, find out set O_candIn be not centainly q Reverse Nearest pair As, and mark；

Step 6：Traverse set S_temp, successively to set O_candIn each unmarked object o carry out K-NN search, Obtain respective Candidate Set nn_o；

Step 7：Successively to set O_candIn each unmarked object o construct respective probability calculation list Sn_o, Sn_o For nn_oWith nn_qUnion and exclude o and q；

Step 8：Set of computations O successively as the following formula_candIn each unmarked object o probability value：

Wherein n_q,oAnd f_q,oIt is the minimum distance and maximum distance of q and o respectively, dist () is the distance between 2 points letters Number, pr { } indicate that certain event is genuine probability；

Step 9：The candidate target that probability value is more than to probability threshold value is inserted into result set and returns.

Preferably, GP rules are specially：As long as o is fully located at the beta pruning region PR of some object co in set S_coIt is interior, Object o is then claimed to meet GP rules, i.e. o is not centainly the Reverse Nearest of q；Wherein beta pruning region PR_coBy in plane to the nearest of q The all the points that distance is more than to object co maximum distances form.In a kind of uncertain monochromatic mutually K-NN search processing method S is set O in step 3_cand, S is set O in step 5_pru。

Compared with the conventional method, technical solution provided by the invention has the advantages that：

First, the uncertain monochromatic mutually K-NN search processing method is in the nearest neighbor search for completing query object q Afterwards, the Reverse Nearest search of q is not carried out directly, but will not be centainly the anti-of q in K-NN search result using GP rules It is excluded to the object of arest neighbors, to further reduce I/O number and time cost on the basis of existing method；

Secondly, the uncertain monochromatic mutually K-NN search processing method utilizes the object formation excluded in previous step Beta pruning region reuses GP rules to further reduce mutual arest neighbors Candidate Set, reduces I/O number and mutual arest neighbors The calculation amount of probability, to reduce time cost；

In addition, the uncertain monochromatic mutually K-NN search processing method is in different uncertain sizes, different data When collection distribution, compared with the conventional method, operational efficiency is high, i.e., I/O costs are low, query time is short；This method is for locating Corresponding query processing problem in LBS is managed, query processing efficiency can be improved.

Description of the drawings

Fig. 1 is the flow chart of uncertain monochromatic mutually K-NN search processing method provided in an embodiment of the present invention；

Fig. 2 is the schematic diagram that nearest neighbor search is carried out on a uncertain data collection, and (a) searches for the uncertain nearest of q Neighbour (b) searches for the uncertain arest neighbors of h, (c) searches for the uncertain arest neighbors of i, (d) searches for the uncertain arest neighbors of c, (e) searches The uncertain arest neighbors of rope e (f) searches for the uncertain arest neighbors of b；

Fig. 3 is the schematic diagram of the R trees for dataset construction shown in Fig. 2；

Fig. 4 is the schematic diagram that GP rule prunings are used on uncertain data collection, (a) beta pruning region PR_h, the beta pruning area (b) Domain PR_i, (c) beta pruning region PR_c, (d) beta pruning region PR_e；

Fig. 5 is the method for the present invention and comparison diagram of the existing method on time cost, (a) on data set LB, (b) in number According on collection CA；

Fig. 6 be the method for the present invention with existing method at I/O time it is several on comparison diagram, (a) on data set LB, (b) in number According on collection CA.

Specific implementation mode

In order to make the purpose , technical scheme and advantage of the present invention be clearer, with reference to the accompanying drawings and embodiments, right The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and It is not used in the restriction present invention.

The description of specific distinct unless the context otherwise, the present invention in element and component, the shape that quantity both can be single Formula exists, and form that can also be multiple exists, and the present invention is defined not to this.Although step in the present invention with label into It has gone arrangement, but is not used to limit the precedence of step, unless expressly stated the order of step or holding for certain step Based on row needs other steps, otherwise the relative rank of step is adjustable.It is appreciated that used herein Term "and/or" one of is related to and covers associated Listed Items or one or more of any and all possible groups It closes.

As shown in Figure 1, a kind of uncertain monochromatic mutually K-NN search processing method, includes the following steps：

Step 1：Initialize most rickle H_rfn, enable set S_temp,O_cand,O_pruFor sky；

Step 2：Traverse R trees, update in the process and obtain all objects to the maximum distance of query object q minimum Value minf, and node and the object beta pruning that minf will be more than to the minimum distance of q, while being keyword by the minimum distance to q It will be preserved to most rickle H by the node of beta pruning and object_rfnIn, fail to be stored in set nn by the object of beta pruning_qIn；

Step 3：By nn_qIn each object o according to its to q minimum distance be keyword be inserted into H_rfnIn；Collecting simultaneously Close O_candIt is upper that object o is judged using GP rules, if o meets beta pruning condition, o is inserted into set O_pru, otherwise by o It is inserted into set O_cand；Moreover, GP rules are specially：As long as o is fully located at O_candIn some candidate target co beta pruning region PR_coInterior, then o is not centainly the Reverse Nearest of q, and o can be excluded, beta pruning region PR_coBy the minimum distance to q in plane More than all the points composition to object co maximum distances；

Step 4：By H_rfnIn element be entirely insertable set S_temp；

Step 5：In set O_pruIt is upper regular using GP, find out set O_candIn be not centainly q Reverse Nearest pair As, and mark；

Step 6：Traverse set S_temp, successively to set O_candIn each unmarked object o carry out K-NN search, Obtain respective Candidate Set nn_o；

Step 7：Successively to set O_candIn each unmarked object o construct respective probability calculation list Sn_o, Sn_o For nn_oWith nn_qUnion and exclude o and q；

Step 8：Set of computations O successively as the following formula_candIn each unmarked object o probability value：

Wherein n_q,oAnd f_q,oIt is the minimum distance and maximum distance of q and o respectively, dist () is the distance between 2 points letters Number, pr { } indicate that certain event is genuine probability；

Step 9：The candidate target that probability value is more than to probability threshold value is inserted into result set and returns.

If by inventor in doctoral thesis《Position does not know mobile object query processing key technology research》It (navigates in Nanjing Empty space flight university, in October, 2013) in a kind of methods based on more wheel nearest neighbor search technologies for proposing, be denoted as MN methods, Steps are as follows：

Step 1：R trees are traversed, the arest neighbors Candidate Set nn of query object q is obtained_q；

Step 2：To nn_qIn each object o, traversal R tree queries obtain the arest neighbors Candidate Set nn of object o_o；

Step 3：To nn_qIn each object o, if q is nn_oIn element, then o is inserted into mutual arest neighbors Candidate Set mnn_q, seek nn_oWith nn_qUnion and exclude o and q, obtain the probability calculation list of o；

Step 4：Mnn is calculated according to the following formula_qIn each object probability value：

Step 5：Judge whether probability value is more than probability threshold value, is to insert objects into result set and return.

If by inventor Patent No. 201610118192.5 patent of invention《In a kind of Instable Space data Monochromatic mutually K-NN search processing method》The method of middle proposition searched for based on arest neighbors and Reverse Nearest, is denoted as the side NR Method, its step are as follows：

Step 1：Initialize most rickle H_rfn, enable set S_tempFor sky；

Step 2：R trees are traversed, the arest neighbors Candidate Set nn of query object q is obtained_q, and by the node of beta pruning during this It is stored in most rickle H with object_rfnIn；

Step 3：By the arest neighbors Candidate Set nn of q_qIn object using the minimum distance to q as keyword be inserted into H_rfnIn；

Step 4：By H_rfnIn element simultaneously be inserted into set S_temp；

Step 5：Traverse H_rfnObtain the Reverse Nearest Candidate Set rnn of q_q；

Step 6：Find out nn_qWith rnn_qMutual arest neighbors Candidate Set mnn of the intersection as q_q；

Step 7：To mnn_qIn each object o, traversal set S_tempThe arest neighbors Candidate Set of o is obtained, nn is denoted as_o；

Step 8：For mnn_qIn each object o construction probability calculation lists Sn_o, Sn_oFor nn_oWith nn_qUnion and exclude o and q；

Step 9：Mnn is calculated according to the following formula_qIn each object o probability value；

Step 10：The candidate target that probability value is more than to probability threshold value is inserted into result set and returns.

By uncertain monochromatic mutually K-NN search processing method provided in an embodiment of the present invention, OptNR methods are denoted as, then Illustrate advantage of the OptNR methods relative to MN methods and NR methods jointly by Fig. 2, Fig. 3 and Fig. 4.

If using MN methods, the K-NN search that R trees carry out q was traversed before this, was not recorded by beta pruning in the process Node and object, according to fig. 2 (a) have：

After executing the step 1, nn_q={ h, i, c, e, b }；

5 traversal R trees are needed in step 2, and are carried out by K-NN search, obtains nn by h, i, c, e, b respectively_h=q, i, e, b, G }, nn_i={ q, h, g, f }, nn_c={ e, d, q }, nn_e={ c, d, b, q }, nn_b={ a, e, h }, is shown in Fig. 2 (b)~(f)；

Step 3 is executed, mnn is obtained_q={ h, i, c, e }, while obtaining Sn_h={ i, c, e, b, g }, Sn_i=h, c, e, b, G, f }, Sn_c={ h, e, i, b, d }, Sn_e={ h, c, i, b, d }；

Step 4 need to be to 4 objects h, i, c, e respectively in probability calculation list Sn_h, Sn_i, Sn_cAnd Sn_eUpper calculating probability.

If using NR methods,：

After executing the step 2, (a) understands nn according to fig. 2_q={ h, i, c, e, b }；In the process by the node of beta pruning and Object is that major key is saved in most rickle H according to the minimum distance to q_rfnIn, then H_rfnIn content be

After executing the step 3,

After executing the step 4, S_temp={ q, h, i, c, e, b, N₅,d,a}；

After executing the step 5, rnn is obtained_q={ h, i, c }；

After executing the step 6, nn is found out_qWith rnn_qMutual arest neighbors Candidate Set mnn of the intersection as q_q, mnn_q=h, i,c}；

After executing the step 7, nn is obtained_h={ q, i, e, b, g }, nn_i={ q, h, g, f }, nn_c={ e, d, q }, is shown in Fig. 2 (b)~(d)；

After executing the step 8, Sn_h={ i, c, e, b, g }, Sn_i={ h, c, e, b, g, f }, Sn_c={ h, e, i, b, d }；

To 3 objects h, i, c respectively in probability calculation list Sn in step 9_h, Sn_iAnd Sn_cUpper calculating probability.

If using OptNR methods,：

The result of step 1~step 2 is consistent with NR methods；

When going to step 3, using GP rules to nn_qAfter object in={ h, i, c, e, b } excludes, according to Fig. 4 (a)~ (c) O is obtained_cand={ h, i, c }, O_pru={ e, b }, H_rfnIn content and method NR in it is identical；

The result of step 4 is identical as NR methods；

Step 5 is in the last round of object set O excluded by GP rules_pruOn, GP rules are reused to set O_candIn Object is judged, according to Fig. 4 (d) it is found that object c is fully located at PR_eIt is interior, it is possible to c to be excluded, then to it into rower Note, obtains O_cand={ h, i, c^*}；

It is only needed to O in step 6_candIn unmarked object h and i carry out K-NN search, obtain nn_h=q, i, e, b, G }, nn_i={ q, h, g, f }；

Step 7 is to h and i construction probability calculation lists Sn_h={ i, c, e, b, g }, Sn_i={ h, c, e, b, g, f }.

It only need to be to 2 objects h and i respectively in probability calculation list Sn in step 8_hAnd Sn_iUpper calculating probability.

In the example shown in Fig. 2~Fig. 4, it is all 1 time that OptNR methods and NR methods access R trees from root node, and MN methods R trees are accessed from root node totally 6 times, and the candidate target number that three kinds of methods obtain is respectively 2,3 and 4.

Since OptNR methods use GP rules in two steps：

For the first time with GP rules to nn_qIn object judged, exclude wherein be unlikely to be q Reverse Nearest pair As, and NR methods are then by traversing H_rfnObtain the Reverse Nearest set rnn of q_qLater again with nn_qIt seeks common ground, it is clear that nn_qIn Object is considerably less than H_rfnThe quantity of middle object and node, because the I/O number and calculation amount of the method OptNR are less than the latter；

In addition, OptNR methods utilize the object set O being excluded_pru, to candidate target collection O_candGP rules are reused, Candidate Set is further reduced, therefore in step 6, the nearest neighbor search for carrying out an object c fewer than NR methods, further Reducing I/O number, while the Candidate Set scale smaller of OptNR methods, the Candidate Set of OptNR methods is { h, i } in this example, and The Candidate Set of NR methods is { h, i, c }, and the time cost that then the former spends in probability calculation also further decreases. Therefore the I/O number and time cost of OptNR methods are minimum in three kinds of methods.

Pair of MN methods, NR methods and OptNR methods on time cost and I/O number is set forth in Fig. 5 and Fig. 6 Than.Obvious OptNR methods are best performances in three kinds of methods.

It is obvious to a person skilled in the art that invention is not limited to the details of the above exemplary embodiments, Er Qie In the case of without departing substantially from spirit or essential attributes of the invention, the present invention can be realized in other specific forms.Therefore, no matter From the point of view of which point, the present embodiments are to be considered as illustrative and not restrictive, and the scope of the present invention is by appended power Profit requires rather than above description limits, it is intended that all by what is fallen within the meaning and scope of the equivalent requirements of the claims Variation is included within the present invention.Any reference signs in the claims should not be construed as limiting the involved claims.

In addition, it should be understood that although this specification is described in terms of embodiments, but not each embodiment is only wrapped Containing an independent technical solution, this description of the specification is merely for the sake of clarity, and those skilled in the art should It considers the specification as a whole, the technical solutions in the various embodiments may also be suitably combined, forms those skilled in the art The other embodiment being appreciated that.

Claims (2)

1. the uncertain monochromatic mutually K-NN search processing method of one kind, which is characterized in that include the following steps：

Step 1：Initialize most rickle H_rfn, enable set S_temp,O_cand,O_pruFor sky；

Step 2：Traverse R trees, update in the process and obtain all objects to the maximum distance of query object q minimum value Minf, and node and the object beta pruning that minf will be more than to the minimum distance of q, while will for keyword by the minimum distance to q It is preserved to most rickle H by the node of beta pruning and object_rfnIn, fail to be stored in set nn by the object of beta pruning_qIn；

Step 3：By nn_qIn each object o according to its to q minimum distance be keyword be inserted into H_rfnIn；Gathering simultaneously O_candIt is upper that object o is judged using GP rules, if o is eligible, o is inserted into set O_pru, otherwise o is inserted into and is collected Close O_cand；

Step 4：By H_rfnIn element be entirely insertable set S_temp。

Step 5：In set O_pruIt is upper regular using GP, find out set O_candIn be not centainly q Reverse Nearest object, and make Label；

Step 6：Traverse set S_temp, successively to set O_candIn each unmarked object o carry out K-NN search, obtain Respective Candidate Set nn_o；

Step 7：Successively to set O_candIn each unmarked object o construct respective probability calculation list Sn_o, Sn_oFor nn_o With nn_qUnion and exclude o and q；

Step 8：Set of computations O successively as the following formula_candIn each unmarked object o probability value：

Wherein n_q,oAnd f_q,oIt is the minimum distance and maximum distance of q and o respectively, dist () is the distance between 2 points functions, pr { } indicates that certain event is genuine probability；

Step 9：The candidate target that probability value is more than to probability threshold value is inserted into result set and returns.

2. a kind of uncertain monochromatic mutually K-NN search processing method according to claim 1, which is characterized in that GP is advised It is specially then：As long as o is fully located at the beta pruning region PR of some object co in set S_coIt is interior, then claim object o to meet GP rules, That is o is not centainly the Reverse Nearest of q；Wherein beta pruning region PR_coIt is farthest to object co by being more than to the minimum distance of q in plane The all the points of distance form.S is set O in the step 3 of right 1_cand, S is set O in step 5_pru。