CN102508640A - Distributed radio frequency identification device (RFID) complex event detection method based on task decomposition - Google Patents

Abstract

The invention discloses a distributed radio frequency identification device (RFID) complex event detection method based on task decomposition. Firstly a complex pattern is decomposed to be a plurality of simple subtasks which are parallelly processed by a plurality of machines to reduce event rate of a single panel point and improve integral throughput; and secondly the distributed RFID complex event detection method provides the complex event detection method based on bitmap indexing when the subtasks are processed to improve detection efficiency.

Description

Distributed RFID complicated event detection method based on the task decomposition

Technical field

The present invention relates to a kind of distributed RFID complicated event detection method.Be specially complex task is decomposed into some simple subtasks,, thereby make the subtask load be lower than the processing node load upper limit, eliminate congested by a plurality of node parallel processings.

Background technology

RF identification (RFID; Radio Frequency Identification) is a kind of non-contact automatic identification technology (AIT; Auto Identification Technology), use radio frequency electromagnetic between reader and labeled mobile article, to transmit data and reach purposes such as identification, tracking.

For the tracking and the analysis of special article etc., need to integrate a plurality of data stream that physically distribute.Yet these carry not too big value of the limited raw data of information, and what people paid close attention to is the high level data with semantic information.It is a kind of technology of special disposal interevent relation that complicated event detects, and it carries out the data matching operation according to user-defined pattern on a plurality of data stream, can aggregate into some simple atomic events the complicated event of physical significance according to rule.Therefore RFID complicated event detection technique is a gordian technique of disposing rfid system.Pattern is the formalized description to complicated event, has defined the atomic event and the constraint condition that constitute complicated event, and a pattern is a task.On macroscopic view, complicated event detects with some atomic event streams as input, through detecting complicated event stream of processing output; On microcosmic, once concrete complicated event detects and can be described as following process: the 1) data in system buffer one time window; 2) the data e that each group is new arrives and triggers the one-time detection action, system searches other from e according to the constraint condition of pattern definition buffered data related data; 3) if find one group of complicated event, then output; 4) new data is inserted the data buffer.

Current, the method that is used for the detection of RFID complicated event mainly contains: 1) complicated event based on finite state machine detects; 2) complicated event based on the Petri net detects; 3) complicated event based on the coupling tree detects.Above method has their own characteristics each on the algorithm that complicated event detects, but all adopts centralized solution in the framework aspect.System runs on the main frame, and same machine accomplished input, detection and output function to entire stream.Yet the arithmetic capability of individual machine is limited, will be absorbed in congested during in the face of high-speed data-flow.On the other hand, centralized solution is difficult to carry out system extension after congested generation, can only the upgrading hardware system, and this just inevitably causes testing process to interrupt.

Along with the development of RFID technology, its application scenarios is just experiencing by simple transformation to complicacy; Moving towards globalization from topical application uses; Combine with the single application system in backstage from RFID, change into and under isomerous environment, realize data sharing etc., this large-scale RFID uses the real-time RFID data stream that will produce flood tide.Under the data stream condition of this flood tide, above-mentioned centralized processing scheme can produce communication and handle congestedly at central processing node, causes the rapid increase of processing delay.Complicated event detects a link of using as RFID, and it receives the data of bottom input and to the upper strata testing result is provided.The popularization that the congested greatly limit RFI D of the processing of this link uses.

Summary of the invention

In order to overcome the congestion problems that prior art produces on the reply high-speed data-flow; The invention provides a kind of distributed RFID complicated event detection method of decomposing: at first based on pattern; Complicated pattern is decomposed some simple subtasks; By many machine parallel processings, thereby the event rate of reduction individual node improves entire throughput; Secondly, when handling the subtask, the present invention proposes a kind of complicated event detection method, improved detection efficiency based on bitmap index.

The technical solution adopted for the present invention to solve the technical problems comprises following content:

1. complex task is decomposed into one group of subtask of satisfying the processing node load limitations, and all subtasks are organized into a task tree.

The present invention takes a kind of iterative strategy to decompose complex task and structure task tree, and concrete steps are following:

1) from candidate events stream when iteration (be atomic event stream complex task for the first time), select several stream to the load upper limit rateT of given processing node, together with the constraint condition between them as the subtask.

2) the selected flow of event of deletion from Candidate Set, and judge whether Candidate Set is empty.If be empty, then iteration finishes; Otherwise execution in step 3).

3) intermediate result that output is detected in the subtask is defined as empty flow of event, matching rate between the residue flow of event in speed and it of calculating the empty flow of event that detects output in the subtask and the Candidate Set, and with empty flow of event adding Candidate Set.Forward step 1) then to, begin to handle to another processing node new round.

Represent the number that candidate events flows with N, (i j) is limited to j on the expression node load, available maximum compression ratio when candidate events stream is [1-i] with m; N dimensional vector X representes selection scheme, and the i position of X is that 1 expression candidate events stream i is selected, and it is selected to be that 0 expression does not have, and (i, j) (i j) sets up corresponding selection vector to X with M in expression.Above-mentioned steps 1) select the method for flow of event to be described below in:

A) to the speed rate of rateT with each candidate events stream _i, carry out the equal proportion scaling and round up processing, become integer to them, round-off error is controlled in 5%, and i representes candidate events stream numbering, rate _iExpression is numbered the speed of the candidate data stream of i.

B) to being numbered 1 candidate events stream,, investigate each j: as j＜rate from 1 to rateT with the load capacity that j representes processing node ₁The time, (1, j)=0, (1, j) the 1st is 0 to X to put m; As j>=rate ₁The time, (1, j)=1, (1, j) the 1st is 1 to X to put m.

C) to being numbered the candidate events stream of i, if i＞N then forwards step f) to; Otherwise i is investigated each j from 1 to rateT: as j＜rate _iThe time, put m (i, j)=m (i-1, j), X (i, j)=X (i-1, j); As j>=rate _iThe time forward step d) to.

D) at selection scheme X (i, j-rate _i) in add i candidate events stream, and calculate its ratio of compression compratio, compare then compratio and m (i-1, size j): if compratio is big, m [i] [j]=compratio then, put X (i, j)=X (i-1, j-rate _i), X (i, j) the i position is 1, otherwise puts m [i] [j]=m [i-1] [j], X (i, j)=X (i-1, j).

E) flow of event i has been investigated all j after, investigate flow of event i+1, put 0 to j, forward step c) to.

F) selection course finishes, vectorial X (N, rateT) (this variable and X (i, j-rate _i) implication identical, what bracket was inner all is subscript, promptly the i value is N, the j value is rateT; The m of hereinafter (N, rateT) in like manner) is selection scheme, corresponding ratio of compression be m (N, rateT).

Above-mentioned steps d) ratio of compression compratio is defined as the ratio of total speed and the speed ratef of the empty flow of event of its detection output of the incoming event stream of subtask in, $\mathrm{Ratef}={\mathrm{Rate}}_1*\underset{k=1}{\overset{M-1}{\mathrm{\Π}}}{\mathrm{\η}}_{(k,k+1)},$ $\mathrm{Compratio}=\frac{\underset{k=1}{\overset{M}{\mathrm{\Σ}}}{\mathrm{Rate}}_k}{\mathrm{Ratef}},$ Wherein M representes the number of selected candidate events stream, rate ₁Represent the speed of the flow of event that first is selected, η representes matching rate, η _{(k, k+1)}Represent the matching rate between adjacent two flows of event.

Above-mentioned steps 3) in empty flow of event and the Candidate Set between the residue flow of event computing method of matching rate following: represent the numbering of selected flow of event with s, the number of selected flow of event is M; The numbering of representing not selected flow of event with h; Represent empty flow of event with f; The matching rate of h and f then

The matching rate of f and h ${\mathrm{\η}}_{f,h}=\frac{{\mathrm{\η}}_{h,f}*\mathrm{RateOut}}{{\mathrm{Rate}}_h}.$

2. adopt based on the complicated event detection algorithm of bitmap index and handle the subtask.

Complicated event detection algorithm based on bitmap index is on the basis of streamjoin algorithm, to add bitmap index to preserve intermediateness, improves detection efficiency.If the subtask comprises P incoming event stream, then specifically detected and claimed to comprise following steps:

1) creates the buffer container of P B+ tree as incoming event stream; By the matching rate ascending order is the incident entity appointment looked-up sequence of different event stream.

Trigger the one-time detection action when 2) each incident entity e arrives:

A) for e sets up P position bitmap, that position 1 that the numbering of flow of event under it is corresponding;

B) in buffer container, search the event matching entity by looked-up sequence.To arbitrary container, in this container, search and e event matching entity, if find, execution in step c), otherwise execution in step d);

C) merge the bitmap of the e and the incident entity that is complementary with it, and, between these two incident entities, set up pointer then with their bitmaps separately of amalgamation result renewal.Judge whether all positions that merge the back bitmap are 1, if then expression detects one group of result, output result, execution in step d); Otherwise forward step b) to, in next container, search.

D) incident e is inserted corresponding container together with its bitmap, finish this detection, wait for the new events entity.

3) deletion expired incident entity and bitmap thereof from the buffering container, and will with the relevant position 0 of the bitmap of the not out of date incident entity of this incident entity coupling.

The invention has the beneficial effects as follows: the present invention has considered magnanimity property, concurrency and the distributivity characteristics of RFID data comparatively all sidedly; Complex task is decomposed into some simple subtasks; By a plurality of processing node parallel detections; Make the subtask load be lower than the load upper limit of processing node, solved the congestion problems of centralized approach when processing high-speed data flows effectively; On the other hand, make the ratio of compression of subtask maximum, thereby controlled processing node less to greatest extent.Simultaneously, the present invention introduces bitmap index in the streamjoin algorithm, has improved the efficient that handle the subtask.

Below in conjunction with accompanying drawing and instance the present invention is further specified.

Description of drawings

Fig. 1 is the synoptic diagram of constructing the process of tree-like task structure iteratively, and wherein the solid line node representes that the atomic event that relates among the pattern flows, the empty flow of event that broken circle is represented that the subtask of child nodes polymerization generation is arranged and detected output;

Fig. 2 is the process flow diagram based on the complicated event detection algorithm of bitmap index.

Embodiment

The complex task that the present invention proposes decomposes and the method for task tree structure is actually a kind of iteration aggregation strategy: flow as initial Candidate Set with the atomic event that complex task was comprised; Therefrom select somely, aggregate into a subtasks together with the constraint condition between them; Then this task being detected the empty flow of event of exporting (distinguishes in order to flow with atomic event stream and complicated event; The intermediate result that the subtask is detected output is defined as empty flow of event) flow together as new Candidate Set with remaining atomic event; Therefrom select some flows of event, aggregate into new subtask together with the restriction relation between them; Repeating above-mentioned polymerization process only remains till the stream in Candidate Set.Above-mentioned iterative process logically forms one tree, promptly tree-like task structure.Wherein, leaf node is represented the atomic event stream that complex task comprises; Non-leaf node has double implication: a) as far as its expression subtask of its child nodes, handle the flow of event input subtask of child nodes representative; B) as far as its father node, its expression virtual events stream.Export target complicated event stream is detected in the subtask of root node representative.To given RFID complicated event pattern (And < A, B, C, D, E >), the overall flow of the method for the invention is described.

1. complex task is decomposed into one group of subtask of satisfying the processing node load limitations, and all subtasks are organized into a task tree.

Suppose: through after the standardization, the processing power of processing node is 1.8, and the speed of atomic event stream is followed successively by 1.0,0.7; 0.8,0.6,0.6; Matching rate between them is as shown in table 1, and two streams of "/" expression do not have direct matching relationship in the table, do not include calculating in.

Table 1 atomic event coupling counting rate meter

	A	B	C	D	E
						A	1	0.45	/	/	/
B	0.57	1	0.5	/	/
						C	/	0.44	1	0.6	/
D	/	/	0.8	/	0.82
						E	/	/	/	0.82	1

The process that iteration generates the subtask is following:

Iteration for the first time: comprise flow of event < A, B, C, D, E>in the Candidate Set, its numbering is followed successively by < 1,2,3,4,5 >, the speed of each flow of event through scaling with round off after be followed successively by < 10,7,8,6,6, >, the value of rateT is 18, the value of N is 5.The implementation of flow of event selection algorithm is following: (below provided the operational process of a selection algorithm, as an example)

1) for the flow of event that is numbered 1, rate ₁=10, thus m (1,1)=m (1,2)=...=m (1,9)=0, X (1,1)=X (1,2)=...=X (1,9)=[00000]; M (1,10)=m (1,11)=...=m (1,18)=1, X (1,10)=X (1,11)=...=X (1,18)=[10000].

2) for the flow of event that is numbered 2, rate ₂=7, so m (2,1)=m (1,1), m (2,2)=m (1,2) ... M (2,6)=m (1,6), X (2,1)=X (1,1), X (2,2)=X (1,2) ... X (2,6)=X (1,6); When 10＞j>=7, the value of compratio is 1, and m (1, value j) is 0, thus m (2, j)=compratio, X (2, j)=[01000]; When 17＞j>=10, the value of compratio is 1, and m (1, value j) is 1, thus m (2, j)=m (1, j), X (2, j)=X (1, j); When j>=17, the value of compratio is 3.78, and m (1, value j) is 1, thus m (2, j)=compratio, X (2, j)=[11000].

3) for the flow of event that is numbered 3, rate ₃=8, so m (3,1)=m (2,1), m (3,2)=m (2,2) ... M (3,7)=m (2,7), X (3,1)=X (2,1), X (3,2)=X (2,2) ... X (3,7)=X (2,7); When 15＞j>=8, the value of compratio is 1, and m (2, value j) is 1, thus m (3, j)=m (2, j), X (3, j)=X (2, j); When 17＞j>=15, the value of compratio is 4.28, and m (2, value j) is 1, thus m (3, j)=compratio, X (3, j)=[01100]; When j>=17, the value of compratio is 4.28, and m (2, value j) is 3.78, thus m (3, j)=compratio, X (3, j)=[01100].

4) for the flow of event that is numbered 4, rate ₄=6, so m (4,1)=m (3,1), m (4,2)=m (3,2) ... M (4,5)=m (3,5), X (4,1)=X (3,1), X (4,2)=X (3,2) ... X (4,5)=X (3,5); When 14＞j>=6, the value of compratio is 1, and m (3, value j) is 1, thus m (4, j)=m (3, j), X (4, j)=X (3, j); When 15＞j>=14, the value of compratio is 2.92, and m (3, value j) is 1, thus m (4, j)=compratio, X (4, j)=[00110]; When j>=15, the value of compratio is 2.92, and m (3, value j) is 4.28, thus m (4, j)=m (3, j), X (4, j)=X (3, j).

5) for the flow of event that is numbered 5, rate ₅=6, so m (5,1)=m (4,1), m (5,2)=m (4,2) ... M (5,5)=m (4,5), X (5,1)=X (4,1), X (5,2)=X (4,2) ... X (5,5)=X (4,5); When 12＞j>=6, the value of compratio is 1, and m (4, value j) is 1, thus m (5, j)=m (4, j), X (5, j)=X (4, j); When 14＞j>=12, the value of compratio is 2.44, and m (4, value j) is 1, thus m (5, j)=compratio, X (5, j)=[00011]; When 15＞j>=14, the value of compratio is 2.44, and m (4, value j) is 2.92, thus m (5, j)=m (4, j), X (5, j)=X (4, j); When j>=15, the value of compratio is 2.44, and m (4, value j) is 4.28, thus m (5, j)=m (4, j), X (5, j)=X (4, j).

Through the selection algorithm screening; Obtaining the result is: X (5; 18)=[01100], B, C are obtained being lower than the maximum compression ratio 4.28 of the load upper limit during as the subtask, this task actual loading 15; Its rate estimation value that detects the empty flow of event F of output is 3.5, and the matching rate between other flow of event and F is as shown in table 2.

Table 2 atomic event coupling counting rate meter

	A	F	D	E
					A	1	0.225	/	/
F	0.642	1	0.801	/
					D	/	0.467	1	0.82
E	/	/	0.82	1

Iteration for the second time: comprise flow of event < A, F, D, E>in the Candidate Set, its numbering is followed successively by < 1,2,3,4 >, the speed of each flow of event through scaling with round off after be followed successively by < 100,35,60,60 >, the value of rateT is 180, the value of N is 4.Through the selection algorithm screening; Obtaining the result is: X (4; 180)=[0111], F, D, E are obtained being lower than the maximum compression ratio 6.74 of the load upper limit during as the subtask, this task actual loading 155; Its rate estimation value that detects the empty flow of event G of output is 22.9, and the matching rate between other flow of event and G is as shown in table 3.

Table 3 atomic event coupling counting rate meter

	A	G
			A	1	0.147
G	0.642	1

Iteration for the third time: comprise flow of event < A, G>in the Candidate Set, its numbering is followed successively by < 1,2 >, the speed of each flow of event through scaling with round off after be followed successively by < 100,22 >, be limited to 180 in the load of processing node.Through selection algorithm screening, obtain the result and be: X (2,180)=[11], A, G are obtained being lower than the maximum compression ratio 8.36 of the load upper limit during as the subtask, this task actual loading 122, its rate estimation value that detects the empty flow of event H of output is 14.7.

When iteration completion for the third time, Candidate Set is empty, and whole iterative process finishes.Fig. 1 has demonstrated iterative process, and B wherein, C form subtask And < B, C>at the 4th layer, and the empty flow of event F of output is detected in this subtask; F, D and E belong to the 3rd layer, form subtask And < F, D, E >, and the empty flow of event G of output is detected in this subtask; G and A belong to the second layer, form subtask And < A, G >, detect export target complicated event H behind this subtask.Three subtasks after the decomposition are by three node parallel processings.

2. adopt complicated event detection algorithm to carry out the subtask processing at processing node based on bitmap index.

Processing node moves centralized detection algorithm and handles the subtask.Because computational resource is wasted in the frequent expired event handling meeting of carrying out, thereby the present invention adopts regularly, and trigger mechanism carries out expired processing.Be elaborated for example with subtask And < F, D, E >:

(1) initialization link: set up three B+ tree respectively as the buffer container of flow of event F, D and E; The bitmap of F, D, E is respectively [100], [010], [001], and the bitmap of son termination fruit is < 111 >; Be that three types of atomic events are set up the detection sequence, be respectively: F → D → E, D → E → F and E → F → D; The regularly triggering cycle is set; Start communication service and receive incoming event; Detect thread then and be absorbed in waiting status.

(2) detect output element: when the entity of certain D class atomic event arrives, earlier it is deposited in cell therefor and sets up bitmap < 010>for it.Go to search the E class incident that is complementary according to detecting sequence then; Suppose to find; Then the bitmap merging (or operation) with corresponding D, E incident entity is < 011 >; And search the corresponding matched incident along the buffer container that detection proceeds to F in proper order, and suppose not find, then this detects action and stops; When the F incident of coupling finally arrives, detect sequence according to it and go to search the D incident that is complementary, this D bitmap is corrected for < 011 >, arrives dbjective state < 111>through after the one query, detects successfully.

(3) expired processing links: after the triggering cycle arrived, the storage container that reverse scanning is all was deleted expired atomic event, and revised the bitmap of the atomic event that matches.In the example of step (2),, then delete the E incident, and the bitmap of D incident is reduced to < 010>if the E incident is expired before the F event comes.

Claims (5)

1. a distributed RFID complicated event detection method of decomposing based on task is characterized in that comprising the steps:

The first step is decomposed into one group of subtask of satisfying the processing node load limitations with complex task, and all subtasks are organized into a task tree;

The present invention takes a kind of iterative strategy to decompose complex task and structure task tree, and concrete steps are following:

1) from candidate events stream, select several stream to the load upper limit rateT of given processing node, together with the constraint condition between them as the subtask;

2) the selected flow of event of deletion from Candidate Set, and judge whether Candidate Set is empty, if be empty, then iteration finishes; Otherwise execution in step 3);

3) intermediate result that output is detected in the subtask is defined as empty flow of event, matching rate between the residue flow of event in speed and it of calculating the empty flow of event that detects output in the subtask and the Candidate Set, and with empty flow of event adding Candidate Set; Forward step 1) then to, begin to handle to another processing node new round;

Second step, adopt based on the complicated event detection algorithm of bitmap index and handle the subtask, establish the subtask and comprise P incoming event stream, then specifically detected and claimed to comprise following steps:

1) creates the buffer container of P B+ tree as incoming event stream; By the matching rate ascending order is the incident entity appointment looked-up sequence of different event stream;

Trigger the one-time detection action when 2) each incident entity e arrives, comprise following content:

A) for e sets up P position bitmap, that position 1 that the numbering of flow of event under it is corresponding;

B) in buffer container, search the event matching entity by looked-up sequence,, in this container, search and e event matching entity to arbitrary container, if find, execution in step c), otherwise execution in step d);

C) merge the bitmap of the e and the incident entity that is complementary with it, and, between these two incident entities, set up pointer then with their bitmaps separately of amalgamation result renewal; Judge whether all positions that merge the back bitmap are 1, if then expression detects one group of result; The output result, execution in step d); Otherwise forward step b) to, in next container, search;

D) incident e is inserted corresponding container together with its bitmap, finish this detection, wait for the new events entity; 3) deletion expired incident entity and bitmap thereof from the buffering container, and will with the relevant position 0 of the bitmap of the not out of date incident entity of this incident entity coupling.

2. distributed RFID complicated event detection method of decomposing based on task according to claim 1 is characterized in that: described first step step 1) for the first time during iteration candidate's flow of event be the atomic event stream in the complex task.

3. distributed RFID complicated event detection method of decomposing according to claim 1 based on task; It is characterized in that: described first step step 1) is represented the number of candidate events stream with N; With m (i; J) be limited to j on the expression node load, available maximum compression ratio when candidate events stream is [1-i]; N dimensional vector X representes selection scheme, and the i position of X is that 1 expression candidate events stream i is selected, and it is selected to be that 0 expression does not have, X (i, j) expression and M (i j) sets up corresponding selection vector, selects the method for flow of event may further comprise the steps:

A) to the speed rate of rateT with each candidate events stream _i, carrying out the equal proportion scaling and become integer with rounding up to handle, round-off error is controlled in 5%, and i representes candidate events stream numbering, rate _iExpression is numbered the speed of the candidate data stream of i;

B) to being numbered 1 candidate events stream,, investigate each j: as j＜rate from 1 to rateT with the load capacity that j representes processing node ₁The time, (1, j)=0, (1, j) the 1st is 0 to X to put m; As j>=rate ₁The time, (1, j)=1, (1, j) the 1st is 1 to X to put m;

C) to being numbered the candidate events stream of i, if i＞N then forwards step f) to; Otherwise i is investigated each j from 1 to rateT: as j＜rate _iThe time, put m (i, j)=m (i-1, j), X (i, j)=X (i-1, j); As j>=rate _iThe time forward step d) to;

D) at selection scheme X (i, j-rate _i) in add i candidate events stream, and calculate its ratio of compression compratio, compare then compratio and m (i-1, size j): if compratio is big, m [i] [j]=compratio then, put X (i, j)=X (i-1, j-rate _i), X (i, j) the i position is 1, otherwise puts z [i] [j]=m [i-1] [j], X (i, j)=X (i-1, j);

E) flow of event i has been investigated all j after, investigate flow of event i+1, put 0 to j, forward step c) to;

F) selection course finishes, vectorial X (N rateT) is selection scheme, corresponding ratio of compression be m (N, rateT).

4. distributed RFID complicated event detection method of decomposing according to claim 1 based on task; It is characterized in that: ratio of compression compratio is defined as the ratio of total speed and the speed ratef of the empty flow of event of its detection output of the incoming event stream of subtask in the described first step step d)

Wherein M representes the number of selected candidate events stream, rate ₁Represent the speed of the flow of event that first is selected, η representes matching rate, η _{(k, k+1)}Represent the matching rate between adjacent two flows of event.

5. distributed RFID complicated event detection method of decomposing according to claim 1 based on task; It is characterized in that: in the described first step step 3) in empty flow of event and the Candidate Set between the residue flow of event computing method of matching rate following: represent the numbering of selected flow of event with s, the number of selected flow of event is M; The numbering of representing not selected flow of event with h; Represent empty flow of event with f; The matching rate of h and f then ${\mathrm{\&eta;}}_{h,f}=\underset{s=1}{\overset{M}{\mathrm{\&Pi;}}}({\mathrm{\&eta;}}_{h,s}*\frac{{\mathrm{Rate}}_f}{{\mathrm{Rate}}_s}),$ The matching rate of f and h ${\mathrm{\&eta;}}_{f,h}=\frac{{\mathrm{\&eta;}}_{h,f}*\mathrm{RateOut}}{{\mathrm{Rate}}_h}.$

Images