CN105722129A - Wireless sensing network event detection method and system based on FSAX-MARKOV model - Google Patents

Abstract

The invention discloses a wireless sensing network event detection method and a system based on an FSAX-MARKOV model. The method comprises following steps of S1: acquiring original monitoring data so as to form a time sequence and using the time sequence as a training sample sequence; S2: according to the training sample sequence, calculating a transition possibilitymatrix of an FSAX symbol sequence; S3: a raining stage: calculating the transition possibility matrix of the FSAX symbol sequence and determining a normal transition possibility threshold valuedeltath of the FSAX symbol sequence; S4: a detection stage: calculating the transition possibility delta of the FSAX symbol sequence of monitoring data in the current sliding window; and S5: according to the deltath and the delta, carrying out abnormal event detection on the monitoring data in the current sliding window. According to the invention, the method and the system are quite high in event detection precision and quite low in false alarm rate; abnormal sections in a sequence can be precisely found; timeliness and reliability of abnormal event detection of the wireless sensing network are improved; and energy and communication bandwidth of the wireless sensing network are largely saved.

Description

A kind of wireless sense network event detecting method based on FSAX-MARKOV model and System

Technical field

The present invention relates to technology of Internet of things field, particularly relate to a kind of wireless sense network based on FSAX-MARKOV model Event detecting method and system.

Background technology

Wireless sense network (Wireless Sensor Networks is called for short WSN) is the dominant form of future network development, And become one new science research field in this century.Propose many urgent need in rationale and two aspects of engineering to solve Problem certainly.Wireless sense network is with low cost, low-power consumption, extensive MANET；Sensor node compact, battery power, Dispose flexibly；And can adapt to monitor the adverse circumstances that manpower is difficult to arrive；These features make wireless sense network greatly Improve the monitoring capacity of disaster prevention.In order to monitor various contingent accident in time (such as landslide, air dirt Dye, forest fire etc.), it is necessary to pay close attention to the abnormal measures that sensor node collects.Therefore, the abnormal number of detection real-time and accurately According to, and early warning particular event, tool is of great significance.

The accident detection technology of wireless sense network sums up and is broadly divided into two classes: 1) some method for detecting abnormality.Point If abnormal i.e. sensing data exceedes certain threshold value of setting, then it is assumed that event occurs.This method is only suitable on a small scale, short-term Single incident monitoring task.2) pattern method for detecting abnormality.In some long-term gradient environment are monitored, paroxysmal complicated thing Part is often difficult to be reported to the police by transfiniting of specified attribute threshold value, it is impossible to describes by simple threshold value, but can regard one as Pattern (event schema), because using mode identification technology to carry out abnormality detection.At present, major part pattern method for detecting abnormality is all It is to carry out on acquired original data space, the data of sensor node collection is not carried out any conversion, although this side Method have certain accuracy of detection.But method is computationally intensive, poor fault tolerance, and energy-saving effect is limited.Can be after processing through overcompression Data space on carry out abnormality detection？Further, in wireless sense network, accident detection technology also to face two and mainly chooses War: 1) accuracy of detection.Owing to being affected by various faults in environment noise and network, sensor node often gives the prison made mistake Measured value, this will certainly have influence on the reliability of accident detection.Therefore, detection method must have fault-tolerance.2) energy has Effect property.Sensor node has very limited amount of energy reserve, and the network lifetime of wireless sense network event monitoring depends on joint Point energy consumption, thus detection method must have energy saving.

In extensive long-term deployment wireless sense network, thousands of sensor node produces the Dimension Time Series of magnanimity, These data contain substantial amounts of redundancy and conceals the dependency of important relationship, if in these original data space directly Carrying out abnormality detection, the great expense incurred of its energy and communication bandwidth will shorten network lifecycle, even makes wireless sense network Monitoring task can not be completed.Therefore, before data are sent to gateway, it is compressed (or dimensionality reduction) be very important.? In event monitoring type WSN application system, quickly identifying that anomalous event is its primary goal from the Monitoring Data of network, it is important Property is even more than Monitoring Data itself.Excavate the temporal correlation between node by data compression method, disappear to greatest extent Redundancy between divisor evidence, while ensureing volume of transmitted data is greatly lowered, is maintained to high-precision event Monitoring Performance, and from magnanimity flow data, extract potential useful information, pattern and trend.

Summary of the invention

The technical problem to be solved in the present invention is for accuracy of detection in prior art the highest, and the defect that energy consumption is big, There is provided one ensure volume of transmitted data is greatly lowered while, be maintained to high-precision event monitoring performance based on The wireless sense network event detecting method of FSAX-MARKOV model and system.

The technical solution adopted for the present invention to solve the technical problems is:

The present invention provides a kind of wireless sense network event detecting method based on FSAX-MARKOV model, including following step Rapid:

S1, gathering primary monitoring data according to constituting time series X as training sample sequence, and it is long to arrange detection window Degree, FSAX symbol sebolic addressing length, glossary of symbols and size thereof；

S2, training sample sequence X is normalized obtains sequence Y ∈ Rⁿ, normalization sequence Y is compressed fall Dimension obtains compressed sequenceCalculate compressed sequenceIn the variance of each sequential subsegment, by average and the side of each sequential subsegment Difference symbolization, it is thus achieved that FSAX symbol sebolic addressing, and calculate FSAX symbol sebolic addressing transition probability matrix；

S3, training stage: calculate the transfer of FSAX symbol sebolic addressing according to the FSAX symbol sebolic addressing transition probability matrix obtained general Rate, and determine normal transition probability threshold value δ of FSAX symbol sebolic addressing_th；

S4, detection-phase: the FSAX symbol sebolic addressing transition probability δ of Monitoring Data in calculating current sliding window mouth；

S5, according to δ_thWith δ, Monitoring Data in current sliding window mouth is carried out accident detection, if certain FSAX symbol sequence Column jump probability is less than δ_th, then judge that monitored area has anomalous event to occur.

Further, in step S2 of the present invention, wireless sensing net node gathers primary monitoring data composition time series X ={ x₁,x₂,…,x_nAs training sample sequence, training sample sequence X is normalized and obtains sequence Y={y₁, y₂…,y_i,…y_nComputing formula be:

$y_i=\frac{x_i-\mathrm{\μ}}{\mathrm{\σ}}$

Wherein, μ is the average of sequence X, and σ is the standard deviation of sequence X.

Further, normalization sequence is compressed by step S2 of the present invention method particularly includes:

To normalization sequence Y ∈ RⁿIt is compressed dimensionality reduction and obtains the compressed sequence of m dimension And m ＜ ＜ n,Computing formula be:

${\stackrel{\‾}{c}}_i=\frac{m}{n}\underset{j=\frac{n}{m}(i-1)+1}{\overset{\frac{n}{m}i}{\Σ}}{y}_j$

Wherein,Represent the i-th sequential subsegment average of normalization sequence Y, by the average of data in each sequential subsegment Represent the data of this subsegment, thus realize data compression to reject redundancy and smooth noise in original series.

Further, step S2 of the present invention calculates compressed sequenceIn the formula of each sequential subsegment variance be:

${\mathrm{\σ}}_i=\frac{m}{n}\underset{k=1}{\overset{\frac{n}{m}}{\Σ}}(y_{i_k}-{\stackrel{\‾}{c}}_i)$

Wherein,Represent the i-th sequential subsegment average of normalization sequence Y.

Further, average and variance are carried out the method for symbolization by step S2 of the present invention particularly as follows:

By compressed sequenceIn the average of each sequential subsegment and variance symbolization, divide according to the normal state corresponding to character set Σ The look-up table of cloth intervals of equal probability division points, chemical conversion is had two-component symbolic vector by each sequential subsegment, it is thus achieved that FSAX symbol Sequence represents that wireless sensing net node gathers primary monitoring data.

Further, step S2 of the present invention fall into a trap calculation symbol sebolic addressing transition probability matrix formula be:

Ω=(τ (S_M,γ))

Wherein, τ (S_M, γ) and it is transition probability, represent in pattern of symbol S_MThe probability of symbol γ, τ (S occur afterwards_M,γ) Computing formula be: τ (S_M, γ) and=P (S_i+1=γ | S_(i-M+1,i)=S_M)；

Wherein, S_iRepresent symbol sebolic addressing S_MIn i-th symbol.

Further, in step S3 of the present invention sequence of calculation transition probability method particularly as follows:

Obtain normal transition probability threshold value δ of the FSAX symbol sebolic addressing of training stage_thIf, FSAX symbol sebolic addressingDefining its M rank markovian mode shifts probability is:

$P\left(S\right)=P\left(S_M\right)\underset{i=1}{\overset{k}{\Π}}\left(\mathrm{\τ}\right(S_{(i,i=M-1)},S_{i+M}\left)\right)$

Wherein, P (S_M) intermediate scheme S_MProbability.

The present invention provides a kind of wireless sense network event detection system based on FSAX-MARKOV model, including:

Sample collection unit, is used for gathering primary monitoring data according to composition time series X as training sample sequence, and sets Put detection window length, FSAX symbol sebolic addressing length, glossary of symbols and size thereof；

Data Computation Unit, obtains sequence Y ∈ R for being normalized training sample sequence Xⁿ, to normalization Sequence Y is compressed dimensionality reduction and obtains compressed sequenceCalculate compressed sequenceIn the variance of each sequential subsegment, by each sequential The average of subsegment and variance symbolization, it is thus achieved that FSAX symbol sebolic addressing, and calculate FSAX symbol sebolic addressing transition probability matrix；

Training unit, general for calculating the transfer of FSAX symbol sebolic addressing according to the FSAX symbol sebolic addressing transition probability matrix obtained Rate, and determine normal transition probability threshold value δ of FSAX symbol sebolic addressing_th；

Detector unit, the FSAX symbol sebolic addressing transition probability δ of Monitoring Data in calculating current sliding window mouth；

Anomalous event identifying unit, for according to δ_thWith δ, Monitoring Data in current sliding window mouth is carried out anomalous event inspection Survey, if certain FSAX symbol sebolic addressing transition probability is less than δ_th, then judge that monitored area has anomalous event to occur.

The beneficial effect comprise that: wireless sense network event based on the FSAX-MARKOV model inspection of the present invention Survey method, sets up symbol sebolic addressing transition probability matrix by the training stage, determines symbol sebolic addressing normal transition probability threshold value, detection Stage calculates the transition probability of current sliding window mouth internal symbol sequence, and decisions making testing result；The method has higher Event detection precision and relatively low rate of false alarm；Can find that the exception in sequence is interval exactly, improve wireless sense network inspection Survey promptness and the reliability of anomalous event, significantly save wireless sense network energy and communication bandwidth, it is possible to increase wireless biography Sense net node searching efficiency and the ability of location abnormal data.

Accompanying drawing explanation

Below in conjunction with drawings and Examples, the invention will be further described, in accompanying drawing:

Fig. 1 is the flow process of the wireless sense network event detecting method based on FSAX-MARKOV model of the embodiment of the present invention Figure；

Fig. 2 is the sequence symbol of the wireless sense network event detecting method based on FSAX-MARKOV model of inventive embodiments Number change normal distribution intervals of equal probability divide；

Fig. 3 is the normal number of the wireless sense network event detecting method based on FSAX-MARKOV model of inventive embodiments According to sequence X₁(t)；

Fig. 4 is the abnormal number of the wireless sense network event detecting method based on FSAX-MARKOV model of inventive embodiments According to sequence X₃(t)；

Fig. 5 is the normal number of the wireless sense network event detecting method based on FSAX-MARKOV model of inventive embodiments According to sequence Y₁(t)；

Fig. 6 is the abnormal number of the wireless sense network event detecting method based on FSAX-MARKOV model of inventive embodiments According to sequence Y₂(t)；

Fig. 7 is the structural frames of the wireless sense network event detection system based on FSAX-MARKOV model of inventive embodiments Figure；

In figure, 701-sample collection unit, 702-Data Computation Unit, 703-training unit, 704-detector unit, 705- Anomalous event identifying unit.

Detailed description of the invention

In order to make the purpose of the present invention, technical scheme and advantage clearer, below in conjunction with drawings and Examples, right The present invention is further elaborated.Should be appreciated that specific embodiment described herein only in order to explain the present invention, not For limiting the present invention.

As it is shown in figure 1, the wireless sense network event detecting method based on FSAX-MARKOV model of the embodiment of the present invention, Comprise the following steps:

S1, gathering primary monitoring data according to constituting time series X as training sample sequence, and it is long to arrange detection window Degree, FSAX symbol sebolic addressing length, glossary of symbols and size thereof；

S2, training sample sequence X is normalized obtains sequence Y ∈ Rⁿ, normalization sequence Y is compressed fall Dimension obtains compressed sequenceCalculate compressed sequenceIn the variance of each sequential subsegment, by average and the side of each sequential subsegment Difference symbolization, it is thus achieved that FSAX symbol sebolic addressing, and calculate FSAX symbol sebolic addressing transition probability matrix；

S3, training stage: calculate the transfer of FSAX symbol sebolic addressing according to the FSAX symbol sebolic addressing transition probability matrix obtained general Rate, and determine normal transition probability threshold value δ of FSAX symbol sebolic addressing_th；

S4, detection-phase: the FSAX symbol sebolic addressing transition probability δ of Monitoring Data in calculating current sliding window mouth；

S5, according to δ_thWith δ, Monitoring Data in current sliding window mouth is carried out accident detection, if certain FSAX symbol sequence Column jump probability is less than δ_th, then judge that monitored area has anomalous event to occur.

Wireless sensing net node gathers primary monitoring data within n cycle and constitutes time series X={x₁,x₂,…,x_n} As training sample sequence, training sample sequence X is normalized and obtains sequence Y={y₁,y₂…,y_i,…y_nMeter Calculation formula is:

$y_i=\frac{x_i-\mathrm{\μ}}{\mathrm{\σ}}---\left(1\right)$

Wherein, μ is the average of sequence X, and σ is the standard deviation of sequence X.

To normalization sequence Y ∈ RⁿIt is compressed dimensionality reduction and obtains the compressed sequence of m dimension $\stackrel{\‾}{C}=\{{\stackrel{\‾}{c}}_1,{\stackrel{\‾}{c}}_2...,{\stackrel{\‾}{c}}_i,...,{\stackrel{\‾}{c}}_m\}$ And m ＜ ＜ n, whereinFor:

${\stackrel{\‾}{c}}_i=\frac{m}{n}\underset{j=\frac{n}{m}(i-1)+1}{\overset{\frac{n}{m}j}{\Σ}}{y}_j---\left(2\right)$

Wherein,Represent the i-th sequential subsegment average of normalization sequence Y, by the average of data in each sequential subsegment Represent the data of this subsegment, thus realize data compression to reject redundancy and smooth noise in original series.

Calculate compressed sequenceIn each sequential subsegment variances sigma_iFormula be:

${\mathrm{\σ}}_i=\frac{m}{n}\underset{k=1}{\overset{\frac{n}{m}}{\Σ}}(y_{i_k}-{\stackrel{\‾}{c}}_i)---\left(3\right)$

By compressed sequenceIn the average of each sequential subsegment and variance symbolization, divide according to the normal state corresponding to character set Σ The look-up table of cloth intervals of equal probability division points, FSAX symbolism method is using the average of sequence subsegment and variance as describing it Average and degree of scatter.I.e. when carrying out data coding, one symbol having two components of chemical conversion is vowed by each sequential subsegment AmountObtain FSAX symbol sebolic addressing S_MRepresent that wireless sensing net node gathers primary monitoring data and constitutes time series.

Original data sequence X is after the pretreatment that piecewise constant approximation technique carries out dimensionality reduction, if character set size is M, Need M-1 division points.Dividing point set B is:

B=β₁,β₂,…,β_M-1 (4)

As shown in table 1, character set Σ (N is given_α=3～10) the normal distribution intervals of equal probability division points corresponding to, If F (x) is Standard Normal Distribution, division points β need to meet following character:

$\begin{array}{cc}F\left({\mathrm{\β}}_{i+1}\right)-F\left({\mathrm{\β}}_i\right)=\frac{1}{M}& F\left({\mathrm{\β}}_1\right)=\frac{1}{M}\end{array}---\left(5\right)$

Division points look-up table corresponding to table 1 character set Σ

B	3	4	5	6	7	8	9	10
									β1	-0.43	-0.67	-0.84	-0.97	-1.07	-1.15	-1.22	-1.28
β2	0.43	0	-0.25	-0.43	-0.57	-0.67	-0.76	-0.84
									β3		0.67	0.25	0	-0.18	-0.32	-0.43	-0.52
β4			0.84	0.43	0.18	0	-0.14	-0.25
									β5				0.97	0.57	0.32	0.14	0
β6					1.07	0.67	0.43	0.25
									β7						1.15	0.76	0.52
β8							1.22	0.84
									β9								1.28

As in figure 2 it is shown, give symbolic number M=3, there are 2 division points β₁=-0.43, β₂=0.43 schematic diagram.Division points After determining, each piecewise interval, by correspondence only one character, falls into the component c of each piecewise interval_iAnd σ_iWill be by corresponding characterWithReplace, thus be converted to FSAX symbol sebolic addressing.

If glossary of symbols is Σ={ α₁,α₂,...,α_M, wherein symbol arranges (such as a, b, c...) by ascending order, FSAX symbol Sequence isA then FSAX symbolic vectorConversion formula is as follows:

If β_j-1≤c_i＜ β_j (6)

If β_j-1≤σ_i＜ β_j

It is provided with two FSAX symbolic vectorsWithDistance measure between definition FSAX symbolic vector is:

Wherein,The calculating of distance measure is according to table 1 division points look-up table, and the character of FSAX symbolization The size of collection Σ determines.Such as set character set Σ=a, b, c, d},According to table 1 division points Look-up table have dist (a, b)=0, dist (a, d)=1.34, then have

Further, the FSAX symbol sebolic addressing S of two a length of l is defined¹And S²Between distance measure be:

Wireless sensing net node is gathered after data sequence is converted to FSAX symbol sebolic addressing, replace original sequence with symbol sebolic addressing Row carry out abnormality detection.In FSAX symbol sebolic addressing, various symbol combination are as an event schema, transfer reaction between symbol The Changing Pattern of raw monitored value；The different variation tendencies of distinct symbols combination corresponding data, therefore, the symbol that occurrence frequency is high Combination is referred to as normal mode, otherwise, the symbol combination that occurrence frequency is low is referred to as abnormal patterns.

If FSAX symbol sebolic addressingIts symbol numbers is M, defines symbol sebolic addressing S_MM rank Markov Chain tra nsfer probability matrix is:

Ω=(τ (S_M,γ)) (9)

Wherein, τ (S_M, γ) and it is that transition probability represents in pattern of symbol S_MSymbol γ probability occurs afterwards, i.e. has:

τ(S_M, γ) and=P (S_i+1=γ | S_(i-M+1,i)=S_M) (10)

Wherein, S_iRepresent symbol sebolic addressing S_MIn i-th symbol.

Then, defining its M rank markovian mode shifts probability is:

$P\left(S\right)=P\left(S_M\right)\underset{i=1}{\overset{k}{\Π}}\left(\mathrm{\τ}\right(S_{(i,i=M-1)},S_{i+M}\left)\right)---\left(11\right)$

Wherein, P (S_M) intermediate scheme S_MProbability.

In wireless sense network event monitoring is applied, although anomalous event pattern is unknown, but the various quilts of measurand Think that what normal situation but can be relatively easy describes by pattern.The strategy monitoring event in this type of application is first to find out Various normal modes, if detected pattern can not be mated with each normal mode, then judge have anomalous event to occur.Therefore, Use the node symbol sebolic addressing when occurring without event as training sample, obtain the probability transfer of symbol sebolic addressing normal mode Threshold value, node utilizes δ_thDetect whether that anomalous event occurs.

According to above related definition, the present invention proposes the inspection of a kind of wireless sense network event based on FSAX-MARKOV model Survey method.The method utilizes the redundancy of time of monitor value, is symbol by FSAX symbolism method by monitor value compression expression Sequence, and on symbol space, carry out accident detection.Markov Transition Probabilities model and just is obtained by training study Often model probabilities transfer threshold value.The method includes study and two stages of detection, and the study stage sets up symbol sebolic addressing transition probability Matrix, determines normal transition probability threshold value δ of symbol sebolic addressing_th；Detection-phase calculates turning of current sliding window mouth internal symbol sequence Move probability, and testing result is decision making.In another embodiment of the present invention, the step that the method realizes is:

Step one: wireless sensing net node gathers primary monitoring data and constitutes time series X as training sample sequence；If Put detection window length, FSAX symbol sebolic addressing length, glossary of symbols and size thereof；

Step 2: training sample sequence X is normalized and obtains sequence Y ∈ Rⁿ；

Step 3: normalization sequence Y is compressed dimensionality reduction and obtains compressed sequence

Step 4: calculate compressed sequenceIn the variance of each sequential subsegment；

Step 5: by compressed sequenceIn the average of each sequential subsegment and variance symbolization, it is thus achieved that FSAX symbol sebolic addressing；

Step 6: calculate FSAX symbol sebolic addressing transition probability matrix；

Step 7: calculate FSAX symbol sebolic addressing transition probability, determine normal transition probability threshold value δ of FSAX symbol sebolic addressing_th；

Step 8: the FSAX symbol sebolic addressing transition probability δ of Monitoring Data in detection-phase calculating current sliding window mouth；

Step 9: wireless sense network is according to δ_thWith δ, Monitoring Data in current sliding window mouth is carried out accident detection.

In order to evaluate the performance of the present invention, data set Ma_Data and Keogh_Data disclosed in employing two kinds in experiment.Real Test middle basic parameter and include sensor node detection window size W, symbol sebolic addressing length k, glossary of symbols size N_α, and signal Signal to noise ratio snr.Other parameters include data compression ratio η, MARKOV exponent number M.Wherein compression ratio η=1-k/n.Compression ratio η with MARKOV exponent number M can be arranged according to concrete practical situations.

Statistical average event recall rate EDR and event rate of false alarm FRR that use Multi simulation running experiment are examined as balancing method Survey abnormal performance indications.And investigate detection window size n, glossary of symbols size N_αAnd signal to noise ratio snr to method EDR and The impact of FRR, the accuracy of detection of the method for inspection, anti-noise ability and robustness.Event recall rate EDR is defined as that event occurs Time, the number of times of nodal test outgoing event accounts for the ratio of event generation total degree.Event rate of false alarm FRR is defined as occurring without event Time, the number of times of nodal test outgoing event accounts for the ratio of nodal test event total degree.

1) Ma_Data data set

Ma_Data data set is produced by procedure below:

$X_1\left(t\right)=sin\left(\frac{40\mathrm{\π}}{n}t\right)+n\left(t\right)$

$X_2\left(t\right)=sin\left(\frac{40\mathrm{\π}}{n}t\right)+n\left(t\right)+e_1\left(t\right)$

$X_3\left(t\right)=\sin \left(\frac{40\mathrm{\π}}{n}t\right)+n\left(t\right)+e_1\left(t\right)+e_2\left(t\right)$

Wherein, t=1,2 ..., n, n=1200, n (t) be average be 0, standard deviation is 0.1 additive Gaussian noise, e₁(t)、 e₂T () is two anomalous events, e₁(t)) it is defined as follows:

n₁T () meets normal distribution N (0,0.5), e₂T () is defined as follows:

From defined above, data sequence X₁(t) be the cycle be the normal sequence of 60, as it is shown on figure 3, give X₁(t) Oscillogram；X₂T () is at X₁T () with the addition of abnormal e₁(t), X₃T () is at X₂T () with the addition of abnormal e₂T (), in interval [600,620], [820,870] change the character of sequence, and the two interval exists anomalous event, as shown in Figure 4, gives X₃ T () oscillogram, represents abnormal interval in boxed area.

2) Keogh_Data data set

Keogh_Data data set is produced by procedure below:

$Y_1\left(t\right)=\sin \left(\frac{50\mathrm{\π}}{n}t\right)+n\left(t\right)$

$Y_2\left(t\right)=sin\left(\frac{50\mathrm{\π}}{n}t\right)+n\left(t\right)+e_3\left(t\right)$

Wherein, t=1,2 ..., n, n=800, n (t) be average be 0, standard deviation is 0.1 additive Gaussian noise, defines different Ordinary affair part e₃(t) be:

From defined above, data sequence Y₁(t) be a cycle be the normal sequence of 32, as it is shown in figure 5, give Y₁(t) oscillogram；Y₂T () is at Y₁T () with the addition of anomalous event e₃T (), changes the property of sequence in interval [400,432] Matter, as shown in Figure 6, gives Y₁T () oscillogram, represents abnormal interval in red boxes region.

Experiment 1: set SNR=18dB, changes W and N_αValue, wherein W=30,50,70,90, N_α=3～10.Exist respectively EDR and FRR of method of testing on two kinds of data sets, experimental result is as shown in table 2, table 3, table 4 and table 5.

Table 2 sequence X₃Anomalous event e in (t)₁The EDR of (t)

Table 3 sequence X₃Anomalous event e in (t)₂The EDR of (t)

Can be seen that from table 2 and table 3, this method detection sequence X₃E in (t)₁(t) and e₂T the EDR of () has with glossary of symbols N α increases and the trend that increases, and EDR is also had a certain impact by detection window W；Work as N_αWhen=5～10, e₁(t) and e₂(t) There is higher EDR, such as, work as W=90, N_αWhen=7, e₁(t) and e₂T the EDR of () is up to more than 90%.

Table 4 sequence X₃The FRR of anomalous event in (t)

Table 5 sequence Y₂The FRR of anomalous event in (t)

Can be seen that from table 4 and table 5, work as N_αDuring increase, this method detection sequence X₃(t)、Y₂T the FRR on () slightly increases Add.But as a whole, in sequence X₃(t)、Y₂T still there is the relatively low FRR of ratio, even if at W=50, N on ()_αWhen=10, sequence Row X₃T the FRR of () is also less than 3%.When W=90, N α=6, sequence Y₂T the FRR of () also only has 0.45%.

Experiment 2: set N_α=7, change W and SNR value, W=30,50,70,90, SNR=6dB～42dB, value with 6dB stepping, respectively EDR and FRR of method of testing on Ma_Data and Keogh_Data, and compare its variation tendency, experiment knot Fruit is as shown in table 6 to table 10.

Table 6 sequence X₃Anomalous event e in (t)₁The EDR of (t)

Table 7 sequence X₃Anomalous event e in (t)₂The EDR of (t)

Table 8 sequence Y₂Anomalous event e in (t)₃The EDR of (t)

Table 9 sequence X₃The FRR sequence of anomalous event in (t)

Table 10 sequence Y₂The FRR sequence of anomalous event in (t)

Can be seen that from table 6 to table 8, this method is to e₁(t)、e₂(t)、e₃The EDR of (t) along with signal to noise ratio snr reduction and Reduce.As SNR >=18dB, e₁EDR >=70% of (t), e₂EDR >=80% of (t), it is possible to efficiently identify anomalous event e₁ (t)、e₂(t)；To anomalous event e₃For (t), as SNR >=12dB, e₃T EDR >=80% of (), this method has higher EDR, when SNR is < during 12dB, to e₁(t)、e₂(t)、e₃T the recall rate of () is all by decline.

Can be seen that from table 9 and table 10, this method is in sequence X₃(t)、Y₂T the FRR on () is along with the reduction of signal to noise ratio snr And increase, maximum FRR is also less than 0.6%.During monitoring, various noises can disturb data variation rule, causes proper symbol Pattern may be identified as abnormal patterns, or exception symbol pattern is identified as normal mode.When signal to noise ratio is bigger, noise Less on monitor value impact, that method can effectively distinguish noise and event causes unusual fluctuations, therefore have higher EDR and Relatively low FRR.And when signal to noise ratio snr reduces, cause method cannot effectively distinguish noise and abnormal data ripple that event causes Dynamic, the event recall rate of method reduces, and rate of false alarm can rise.Comprehensive Experiment overall condition, method the most still has There are higher event recall rate and extremely low event rate of false alarm, there is good noiseproof feature.

In sum, the method for the present invention utilizes FSAX method the primary monitoring data of sensor node collection to be expressed as Symbol sebolic addressing, on compressed symbolic sequence space, utilizes Markov model to set up the symbol transition probability matrix of FSAX, with Time, utilize unsupervised learning method to obtain the symbol sebolic addressing exception probability threshold value of node.The method of test result indicate that has higher Event detection rate and relatively low rate of false alarm, significantly reduce Node compression extract with identify abnormal information energy consumption, more excellent Compromise energy and monitoring quality.Therefore, the present invention can adapt to event of different nature in abnormality detection, for wireless sense network Early warning technology provides a kind of light weight method.

As it is shown in fig. 7, the wireless sense network event detection system based on FSAX-MARKOV model of the embodiment of the present invention, For realizing the wireless sense network event detecting method based on FSAX-MARKOV model of the embodiment of the present invention, including:

Sample collection unit 701, is used for gathering primary monitoring data according to composition time series X as training sample sequence, And detection window length, FSAX symbol sebolic addressing length, glossary of symbols and size thereof are set；

Data Computation Unit 702, obtains sequence Y ∈ R for being normalized training sample sequence Xⁿ, to normalizing Change sequence Y is compressed dimensionality reduction and obtains compressed sequenceCalculate the variance of each sequential subsegment in compressed sequence C, by time each The average of sequence subsegment and variance symbolization, it is thus achieved that FSAX symbol sebolic addressing, and calculate FSAX symbol sebolic addressing transition probability matrix；

Training unit 703, turns for calculating FSAX symbol sebolic addressing according to the FSAX symbol sebolic addressing transition probability matrix obtained Move probability, and determine normal transition probability threshold value δ of FSAX symbol sebolic addressing_th；

Detector unit 704, the FSAX symbol sebolic addressing transition probability δ of Monitoring Data in calculating current sliding window mouth；

Anomalous event identifying unit 705, for according to δ_thWith δ, Monitoring Data in current sliding window mouth is carried out anomalous event Detection, if certain FSAX symbol sebolic addressing transition probability is less than δ_th, then judge that monitored area has anomalous event to occur.

It should be appreciated that for those of ordinary skills, can be improved according to the above description or be converted, And all these modifications and variations all should belong to the protection domain of claims of the present invention.

Claims (8)

1. a wireless sense network event detecting method based on FSAX-MARKOV model, it is characterised in that include following step Rapid:

S1, gather primary monitoring data according to constitute time series X as training sample sequence, and arrange detection window length, FSAX symbol sebolic addressing length, glossary of symbols and size thereof；

S2, training sample sequence X is normalized obtains sequence Y ∈ Rⁿ, normalization sequence Y is compressed dimensionality reduction and obtains To compressed sequenceCalculate compressed sequenceIn the variance of each sequential subsegment, by average and the variance symbol of each sequential subsegment Number change, it is thus achieved that FSAX symbol sebolic addressing, and calculate FSAX symbol sebolic addressing transition probability matrix；

S3, training stage: calculate FSAX symbol sebolic addressing transition probability according to the FSAX symbol sebolic addressing transition probability matrix obtained, and Determine normal transition probability threshold value δ of FSAX symbol sebolic addressing_th；

S4, detection-phase: the FSAX symbol sebolic addressing transition probability δ of Monitoring Data in calculating current sliding window mouth；

S5, according to δ_thWith δ, Monitoring Data in current sliding window mouth is carried out accident detection, if certain FSAX symbol sebolic addressing turns Move probability less than δ_th, then judge that monitored area has anomalous event to occur.

Wireless sense network event detecting method based on FSAX-MARKOV model the most according to claim 1, its feature exists In, in step S2, wireless sensing net node gathers primary monitoring data composition time series X={x₁,x₂,…,x_nAs training Sample sequence, is normalized training sample sequence X and obtains sequence Y={y₁,y₂…,y_i,…y_nComputing formula For:

$y_i=\frac{x_i-\mathrm{\&mu;}}{\mathrm{\&sigma;}}$

Wherein, μ is the average of sequence X, and σ is the standard deviation of sequence X.

Wireless sense network event detecting method based on FSAX-MARKOV model the most according to claim 2, its feature exists In, normalization sequence is compressed by step S2 method particularly includes:

To normalization sequence Y ∈ RⁿIt is compressed dimensionality reduction and obtains the compressed sequence of m dimension $\stackrel{\&OverBar;}{C}=\{{\stackrel{\&OverBar;}{c}}_1,{\stackrel{\&OverBar;}{c}}_2...,{\stackrel{\&OverBar;}{c}}_i,...,{\stackrel{\&OverBar;}{c}}_m\},$ And m ＜ ＜ n,Computing formula be:

${\stackrel{\&OverBar;}{c}}_i=\frac{m}{n}\underset{j=\frac{n}{m}(i-1)+1}{\overset{\frac{n}{m}i}{\&Sigma;}}{y}_j$

Wherein,Represent the i-th sequential subsegment average of normalization sequence Y, represented by the average of data in each sequential subsegment The data of this subsegment, thus realize data compression to reject redundancy and smooth noise in original series.

Wireless sense network event detecting method based on FSAX-MARKOV model the most according to claim 3, its feature exists In, step S2 calculates compressed sequenceIn the formula of each sequential subsegment variance be:

${\mathrm{\&sigma;}}_i=\frac{m}{n}\underset{k=1}{\overset{\frac{n}{m}}{\&Sigma;}}(y_{i_k}-{\stackrel{\&OverBar;}{c}}_i)$

Wherein,Represent the i-th sequential subsegment average of normalization sequence Y.

Wireless sense network event detecting method based on FSAX-MARKOV model the most according to claim 1, its feature exists In, average and variance are carried out the method for symbolization by step S2 particularly as follows:

By compressed sequenceIn the average of each sequential subsegment and variance symbolization, according to the normal distribution etc. corresponding to character set ∑ The look-up table of probability interval division points, chemical conversion is had two-component symbolic vector by each sequential subsegment, it is thus achieved that FSAX symbol sebolic addressing Represent that wireless sensing net node gathers primary monitoring data.

Wireless sense network event detecting method based on FSAX-MARKOV model the most according to claim 1, its feature exists In, step S2 calculation symbol sebolic addressing transition probability matrix formula of falling into a trap it is:

Ω=(τ (S_M,γ))

Wherein, τ (S_M, γ) and it is transition probability, represent in pattern of symbol S_MThe probability of symbol γ, τ (S occur afterwards_M, γ) meter Calculation formula is: τ (S_M, γ) and=P (S_i+1=γ | S_(i-M+1,i)=S_M)；

Wherein, S_iRepresent symbol sebolic addressing S_MIn i-th symbol.

Wireless sense network event detecting method based on FSAX-MARKOV model the most according to claim 1, its feature exists In, in step S3 sequence of calculation transition probability method particularly as follows:

Obtain normal transition probability threshold value δ of the FSAX symbol sebolic addressing of training stage_thIf, FSAX symbol sebolic addressingDefining its M rank markovian mode shifts probability is:

$P\left(S\right)=P\left(S_M\right)\underset{i=1}{\overset{k}{\&Pi;}}\left(\mathrm{\&tau;}\right(S_{(i,i=M-1)},S_{i+M}\left)\right)$

Wherein, P (S_M) intermediate scheme S_MProbability.

8. a wireless sense network event detection system based on FSAX-MARKOV model, it is characterised in that including:

Sample collection unit, is used for gathering primary monitoring data according to composition time series X as training sample sequence, and arranges inspection Survey length of window, FSAX symbol sebolic addressing length, glossary of symbols and size thereof；

Data Computation Unit, obtains sequence Y ∈ R for being normalized training sample sequence Xⁿ, to normalization sequence Y It is compressed dimensionality reduction and obtains compressed sequenceCalculate compressed sequenceIn the variance of each sequential subsegment, by each sequential subsegment Average and variance symbolization, it is thus achieved that FSAX symbol sebolic addressing, and calculate FSAX symbol sebolic addressing transition probability matrix；

Training unit, for calculating FSAX symbol sebolic addressing transition probability according to the FSAX symbol sebolic addressing transition probability matrix obtained, And determine normal transition probability threshold value δ of FSAX symbol sebolic addressing_th；

Detector unit, the FSAX symbol sebolic addressing transition probability δ of Monitoring Data in calculating current sliding window mouth；

Anomalous event identifying unit, for according to δ_thWith δ, Monitoring Data in current sliding window mouth is carried out accident detection, if Certain FSAX symbol sebolic addressing transition probability is less than δ_th, then judge that monitored area has anomalous event to occur.