CN110213788A - WSN abnormality detection and kind identification method based on data flow space-time characteristic - Google Patents
WSN abnormality detection and kind identification method based on data flow space-time characteristic Download PDFInfo
- Publication number
- CN110213788A CN110213788A CN201910518513.4A CN201910518513A CN110213788A CN 110213788 A CN110213788 A CN 110213788A CN 201910518513 A CN201910518513 A CN 201910518513A CN 110213788 A CN110213788 A CN 110213788A
- Authority
- CN
- China
- Prior art keywords
- space
- model
- state
- time
- transition probability
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 24
- 238000001514 detection method Methods 0.000 title claims abstract description 21
- 230000005856 abnormality Effects 0.000 title claims abstract description 18
- 239000011159 matrix material Substances 0.000 claims abstract description 54
- 230000007704 transition Effects 0.000 claims abstract description 47
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 23
- 239000000284 extract Substances 0.000 claims abstract description 12
- 230000000644 propagated effect Effects 0.000 claims abstract description 11
- 230000002159 abnormal effect Effects 0.000 claims abstract description 9
- 238000004364 calculation method Methods 0.000 claims description 10
- 238000013507 mapping Methods 0.000 claims description 5
- 238000004321 preservation Methods 0.000 claims description 4
- 230000002123 temporal effect Effects 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 230000010354 integration Effects 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 2
- 238000011478 gradient descent method Methods 0.000 claims description 2
- 238000012544 monitoring process Methods 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 6
- 230000002547 anomalous effect Effects 0.000 description 2
- 230000006854 communication Effects 0.000 description 2
- 239000012141 concentrate Substances 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 230000002779 inactivation Effects 0.000 description 1
- 238000003475 lamination Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/06—Testing, supervising or monitoring using simulated traffic
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/08—Testing, supervising or monitoring using real traffic
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W84/00—Network topologies
- H04W84/18—Self-organising networks, e.g. ad-hoc networks or sensor networks
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Testing Or Calibration Of Command Recording Devices (AREA)
Abstract
The present invention relates to WSN abnormality detections and kind identification method based on data flow space-time characteristic, comprising the following steps: step S1: the real-time stream of sensor target node and its neighbor node to be measured is obtained using space-time sliding window;Step S2: the real-time stream that will acquire is mapped to corresponding state space, and is built into the form of Markov chain, calculates state transition probability matrix based on Markov chain and crossing condition transition probability matrix extracts the space-time characteristic of real-time stream;Step S3: building convolutional neural networks models of classifying, and training obtain trained convolutional neural networks model of classifying more more;Step S4: the space-time characteristic of real-time stream is input in trained more classification convolutional neural networks models, by propagated forward, calculates output result;Step S5: it is whether abnormal that data flow is judged according to model output result, and distinguishes failure exception and event anomalies.Real-time exception monitoring and the Exception Type identification of wireless sensor network can be achieved in the present invention.
Description
Technical field
The present invention relates to a kind of WSN abnormality detections and kind identification method based on data flow space-time characteristic.
Background technique
Wireless sensor network (Wireless SensorNetwork, WSN) is by bad weather, natural event, instrument
The influence of the factors such as failure, sensing data happen occasionally extremely.Since anomalous event and instrument failure can lead to sensor
Similar exceptional value is generated, therefore timely and accurately detection sensor exceptional value and distinguish Exception Type, to the control quality of data
It is significant with the abnormal source of judgement.
WSN method for detecting abnormality is the research hotspot of recent domestic.The wherein abnormal inspection based on data space-time characteristic
Survey method and method for detecting abnormality based on classification are the classic algorithms in the field.Method based on data space-time characteristic can be one
Determine to judge data type in degree, but does not combine the temporal characteristics of data and space characteristics, and be overly dependent upon vacation
If data distribution, and assume to tend not to reflect the true distribution of data.Method based on classification not by data distribution and
The limitation that data area is assumed, but since tagged data is limited, classifier cannot catch the exception whole features of data, it is difficult to
Further discriminate between abnormal data type.
Summary of the invention
In view of this, the purpose of the present invention is to provide a kind of WSN abnormality detections and class based on data flow space-time characteristic
Real-time exception monitoring and the Exception Type identification of wireless sensor network are realized in type recognition methods.
To achieve the above object, the present invention adopts the following technical scheme:
A kind of WSN abnormality detection and kind identification method based on data flow space-time characteristic, comprising the following steps:
Step S1: the real-time number of wireless sensor network target node and its neighbor node is obtained using space-time sliding window
According to stream;
Step S2: the real-time stream that will acquire is mapped to corresponding state space, and is built into the form of Markov chain,
State transition probability matrix is calculated based on Markov chain and crossing condition transition probability matrix extracts the space-time spy of real-time stream
Sign;
Step S3: building convolutional neural networks models of classifying more, and sample instruction is extracted from sensor historic data to be measured
Practice convolutional neural networks models of classifying, the trained convolutional neural networks models of classifying obtained more more;
Step S4: the space-time characteristic of real-time stream is input to trained convolutional neural networks pattern die of classifying more
In type, by propagated forward, output result is calculated;
Step S5: it is whether abnormal that data flow is judged according to model output result, and distinguishes failure exception and event anomalies.
Further, the space-time sliding window by the W nearest moment of destination node and neighbor node detection data
Composition obtains the real-time stream of destination node and neighbor node by space-time sliding window model.
Further, the real-time stream that will acquire is mapped to corresponding state space, specifically:
If sensing data sequence is { u1, u2..., ut, utMeasured value for sensor in t moment, utDifference feature
For Δ ut=ut-ut-1;Original series are converted sequence of differences { Δ u by the difference feature of each data first in the sequence of calculation1,
Δu2..., Δ ut, and sensing data sequence is mapped to corresponding state by the size of its difference feature according to 3 σ criterion
Space;State space S includes 9 states, is a, b, c, d, e, f, g, h, k respectively, for any in sensing data sequence
Value utIf utDifference feature Δ utMeet the corresponding conditions in mapping function, then by utBe mapped to state space S=a, b, c,
D, e, f, g, h, k } in corresponding states, specific mapping function are as follows:
Wherein, σ is sequence of differences { Δ u1, Δ u2..., Δ utStandard deviation, sensing data sequence { u1, u2...,
utIt is changed into status switch { s after state maps1, s2..., st}。
Further, described that the temporal characteristics that state transition probability matrix extracts data flow, meter are calculated based on Markov chain
The space characteristics that crossing condition transition probability matrix extracts data flow are calculated, specific as follows:
(1) calculation method of state transition probability matrix is as follows:
A. the destination node status switch { s after state being mapped1, s2..., stIt is configured to single order Markov chain model X
={ X1, X2..., Xt, the state of t moment is only related with the state at t-1 moment in X model.The state space of X model is S=
{ a, b, c, d, e, f, g, h, k };
B. the state transition probability p of X model constructed by destination node status switch is calculatedij, pijIndicate model X in t-1
Moment goes out present condition siAfterwards, go out present condition s in t momentjProbability, wherein si, sj∈S。pijCalculation formula it is as follows:
In formula, N () is used to calculate the total degree of state appearance;
C. it is all possible in state space S={ a, b, c, d, e, f, g, h, k } to calculate destination node status switch
State transition probability pij, state transition probability matrix P is constituted, the size of matrix P is 9 × 9, matrix are as follows:
Wherein, pij>=0 and
(2) calculation method of crossing condition transition probability matrix is as follows:
A. destination node status switch and neighbor node status switch after state being mapped are built into Markov chain respectively
Model;For the neighbor node B of destination node A and node A, the status switch of A, B node is built into Markov Chain respectively
ModelWithThe state space of model is respectively SA=a, b, c, d, e, f,
G, h, k }, SB={ a, b, c, d, e, f, g, h, k }.
B. the crossing condition transition probability of destination node status switch and neighbor node status switch is then calculated
Indicate model XAGo out present condition at the t-1 momentAfterwards, model XBGo out present condition in t momentProbability,
In Calculation formula it is as follows:
Wherein N () is used to calculate the total degree of state appearance;
C. destination node and neighbor node status switch are finally calculated in state space SA=a, b, c, d, e, f, g, h,
k}、SBAll possible crossing condition transition probability in={ a, b, c, d, e, f, g, h, k }Constitute crossing condition transfer
Probability matrix PAB, matrix PABSize be 9 × 9, matrix are as follows:
Wherein,And
Further, the step S3 specifically:
Step S31: building convolutional neural networks models of classifying, model is by input layer, convolutional layer C1, pond layer S1, volume more
Lamination C2, pond layer S2, full articulamentum FC1, full articulamentum FC2 and output layer are constituted, and totally 8 layers, the activation primitive in model is
ReLU function;
Each layer of effect and design parameter are provided that input layer in more classification convolutional neural networks models: input number
According to the space-time characteristic for the space-time sliding window data extracted based on Markov, space-time
Feature is that targeted sites space-time characteristic set of matrices Π, Π are made of n 9 × 9 matrix, including targeted sites
State transition probability matrix Π11With the crossing condition transition probability matrix Π of targeted sites and neighbor site12, Π13...,
Π1n。
Alternate convolution and pond layer: alternate convolutional layer and pond layer, that is, convolutional layer (C1), pond layer (S1), convolutional layer
(C2), pond layer (S2), alternate convolutional layer is from pond layer for extracting space-time characteristic set of matrices Π in different regional areas
Characteristic pattern.Convolutional layer (C1) is 3 × 3 by 64 sizes, and the convolution kernel that stride is 1 × 1 carries out convolution to set of matrices Π
Operation, and output characteristic pattern is obtained by ReLU activation primitive.Pond layer (S1) is 2 × 2 by size, the pond that stride is 1 × 1
Change window to compress the characteristic pattern of extraction, reduces redundancy feature.The convolution that convolutional layer (C2) is 3 × 3 by 128 sizes
Core continues to extract higher level feature.Pond layer (S2) passes through the pond window compressive features that size is that 2 × 2 strides are 1 × 1
Figure, obtains final feature.
Full articulamentum: model uses 2 full articulamentums in a manner of multi-layer artificial neural network to convolutional layer and pond layer
The characteristic pattern finally obtained carries out integration and dimensionality reduction, the neuron number of full articulamentum are respectively 128 and 64, connects entirely by 2
After connecing layer, characteristic pattern is converted into the vector of 64 dimensions.
Output layer: it is connected to output layer entirely by 64 dimensional vectors that Softmax classifier exports full articulamentum, and calculates
The output probability of each classification, the highest classification of output probability are the calculating classification of sample.Output layer includes 4 neurons, generation
The type of entry mark node data stream, respectively normally, event anomalies and stuck-at fault exception and DRIFT TYPE failure exception.Root
It whether abnormal can determine whether data flow according to the calculated result of output layer, and distinguish failure exception and event anomalies.
Step S32: normal, event anomalies, drifting fault and fixed value failure classes are extracted from sensor historic data
Sample, composing training sample set and test sample set;
Step S33: training sample is mapped to corresponding state space, the form of Markov chain is built into, is based on
Markov chain calculates state transition probability matrix and crossing condition transition probability matrix extracts the space-time characteristic of sample;
Step S34: by the space-time characteristic for the training sample being calculated be input to more classification convolutional neural networks models into
Row training, until model convergence, preservation model structural information and model parameter information
Further, the training process of more classification convolutional neural networks models includes that forward and backward is propagated, forward direction
The different characteristic of input layer data is extracted in communication process by convolutional layer and pond layer, full articulamentum integration characteristics pass through
Softmax obtains classification results, and calculates intersection entropy loss, calculates gradient value according to chain rule when back-propagating, by with
Machine gradient descent method updates each layer weight.
Compared with the prior art, the invention has the following beneficial effects:
1, the present invention obtains the real-time stream of sensor, it can be achieved that wireless sensor network using space-time sliding window model
The real-time exception monitoring and Exception Type of network identify.
2, the present invention has been sufficiently reserved the temporal characteristics and space characteristics of sensor data stream, and compared to it is classical based on
The method for detecting abnormality of space-time characteristic, the present invention is during abnormality detection independent of the hypothesis of sensor data stream point
Cloth more meets reality.
3, the present invention devises convolutional neural networks model of classifying 8 layers more and classifies to the space-time characteristic matrix of data flow
Identification, and compared to classical disaggregated model, convolutional neural networks models of classifying can extract more from limited sample data
More information for being conducive to classification, nicety of grading with higher can be effectively detected exception and distinguish Exception Type.
Detailed description of the invention
Fig. 1 is the method flow schematic diagram of the embodiment of the present invention;
Fig. 2 is space-time sliding window mouth mold in one embodiment of the invention.
Specific embodiment
The present invention will be further described with reference to the accompanying drawings and embodiments.
Fig. 1 is please referred to, the present invention provides a kind of based on the WSN abnormality detection of data flow space-time characteristic and type identification side
Method, and implement on the LWSNDR wireless sensor network of SensorScope wireless sensor network and tape label, it is embodied
Mode is as follows:
Step S1: constructing and training convolutional neural networks models of classifying more, the specific steps are as follows:
Step S11: normal, event anomalies, drifting fault and fixed value event are extracted from sensor historic data first
Hinder class sample, data set situation is as shown in table 1;Then the 80% of total sample is randomly choosed from each sample set is used as training sample
This, 20% is test sample.
(1) in SensorScope wireless sensor network case study on implementation, selecting No. 2, No. 7 and No. 9 sensors is target
Node, neighbor node number 4 choose normal, environment using the temperature data that node is collected as detection variable from historical data
Abnormal, drifting fault and fixed value failure classes sample collect 3 sample sets altogether.The sample dimension of SensorScope sample set is
30 × 5, at the time of the row of sample represents different, column represent destination node and its neighbor node.
(2) it in the LWSNDR wireless sensor network case study on implementation of tape label, selects in single-hop wireless sensor network
The sensor node of interior 1 and outdoor No. 4 nodes interfered by hot-water bottle is destination node, undisturbed indoor No. 2 with
Outdoor No. 3 nodes are respective neighbor node, chosen from the humidity (H) and temperature (T) data of collection event anomalies sample with
Normal sample collects 4 sample sets.The sample dimension of LWSNDR sample set is 34 × 2, at the time of the row of sample represents different,
Column represent destination node and its neighbor node.
1 sample set situation of table
Step S12: training sample is mapped to corresponding state space, the form of Markov chain is built into, is then based on
Markov chain calculates state transition probability matrix and crossing condition transition probability matrix extracts the space-time characteristic of training sample.
(1) SensorScope-2, SensorScope-7 are calculated, SensorScope-9 training sample concentrates each sample
State transition probability matrix and crossing condition transition probability matrix.For each sample, state transition probability matrix is by mesh
Mark node state sequence is calculated, and matrix size is 9 × 9;Crossing condition transition probability matrix is by destination node status switch
It is calculated with neighbor node status switch, matrix size is 4 × 9 × 9.Two kinds of matrixes constitute space-time characteristic set of matrices Π,
The data dimension 5 × 9 × 9 of Π, Π are the space-time characteristic of sample.
(2) LWSNDR-1-H, LWSNDR-1-T, LWSNDR-2-H are calculated, LWSNDR-2-T training sample concentrates each sample
This state transition probability matrix and crossing condition transition probability matrix.For each sample, state transition probability matrix by
Destination node status switch is calculated, and matrix size is 9 × 9;Crossing condition transition probability matrix is by destination node state sequence
Column are calculated with neighbor node status switch, and matrix size is 9 × 9.Two kinds of matrixes constitute space-time characteristic set of matrices Π, Π
Data dimension be space-time characteristic that 2 × 9 × 9, Π is sample.
Step S13: more than 8 layers points be made of input layer, alternate convolution and pond layer, full articulamentum and output layer is constructed
Class convolutional neural networks model, the input of model are the space-time characteristic of sample, are exported as the corresponding Exception Type of sample.
Model hyper parameter setting are as follows: learning rate (learning_rate) is 0.001, learning rate dynamic attenuation rate (decay_
It rate) is 0.9, the number of iterations (epoch) is 200, and random inactivation rate (dropout) is 0.2.Due to SensorScope sample
Collect different with the sample size of LWSNDR sample set, thus it is in the implementation case that batch size of SensorScope sample set is big
Small (batch_size) is set as 100, sets 60 for batch size (batch_size) of LWSNDR sample set model.
Step S14: being successively input to more classification convolutional neural networks models for the space-time characteristic of training sample and be trained,
Obtain 7 trained models.
(1) in SensorScope wireless sensor network case study on implementation, by SensorScope-2, SensorScope-
7, SensorScope-9 sample sets are input to more classification convolutional neural networks models and are trained, and obtain trained S2, S7,
S9 model.
(2) in LWSNDR wireless sensor network case study on implementation, by LWSNDR-1-H, LWSNDR-1-T, LWSNDR-2-
H, LWSNDR-2-T sample set are input to more classification convolutional neural networks models and are trained, and obtain trained L1-H, L1-T,
L2-H, L2-T model.
Model training includes the following contents:
(1) propagated forward (Forward).Layer in model successively carries out propagated forward according to sequence from front to back, and L layers
Output be L+1 layers of input.At the end of propagated forward, loss function is defined, for measuring point of network model output
The difference of class result and sample true tag, the implementation case is using cross entropy loss function as loss function.
(2) backpropagation (Backward).First with chain rule and stochastic gradient descent algorithm, calculate from back to front
Weight gradient updates every layer of corresponding weight then in conjunction with learning rate and weight gradient.
(3) loop iteration training network.Pass through successive ignition training convolutional neural networks models of classifying, each iteration packet more
Include propagated forward and backpropagation.Loop iteration to model is restrained, preservation model structural information and model parameter information.
Step S2: the real-time number of wireless sensor network target node and its neighbor node is obtained using space-time sliding window
According to stream.
(1) in SensorScope wireless sensor network case study on implementation, the sensing data of model training will be had neither part nor lot in
Be considered as real time data, using space-time sliding window obtain No. 2, No. 7, the real time temperature number of No. 9 destination nodes and its neighbor node
According to stream, 3 real-time streams are obtained, the dimension of data flow is 30 × 5.
(2) in LWSNDR wireless sensor network case study on implementation, the sensing data for having neither part nor lot in model training is considered as
Real time data obtains the real time temperature number of indoor No. 1 and outdoor No. 4 destination nodes and its neighbor node using space-time sliding window
According to stream and real-time humidity data stream, 4 real-time streams are obtained, the dimension of data flow is 34 × 2.
The space-time sliding window model of building is as follows:
Space-time sliding window is made of the detection data at W nearest moment of destination node and neighbor node, foundation when
Empty sliding window is as shown in Fig. 2, window size is n × W, and row represents different nodes, wherein S1Indicate destination node,
{S2... SnIndicate destination node communication range in neighbor node set;Inspection of the column representative sensor node in different moments
Measured value, { ui1, ui2..., uiWIndicate corresponding node SiDetection sequence within the nearest W moment.When new data generates, when
The data of parent window least significant end are deleted, and new data are added by empty one position of sliding window entirety forward slip, with this reality
The update of current sky sliding window.
Step S3: the real-time stream that will acquire is mapped to corresponding state space, is built into the form of Markov chain, and
State transition probability matrix is calculated based on Markov chain and crossing condition transition probability matrix extracts the space-time spy of real-time stream
Sign;
(1) in SensorScope wireless sensor network case study on implementation, the space-time for the real-time stream being calculated is special
Sign is the data dimension of space-time characteristic set of matrices Π, Π that state transition probability matrix and crossing condition transition probability matrix form
Degree 5 × 9 × 9.
(2) in LWSNDR wireless sensor network case study on implementation, the space-time characteristic for the real-time stream being calculated is
The data dimension 2 of space-time characteristic the set of matrices Π, Π of state transition probability matrix and crossing condition transition probability matrix composition
×9×9。
Step S4: it loads the model of preservation and trained model parameter, the space-time for the real-time stream that will acquire is special
Sign matrix is separately input into corresponding trained classification convolutional neural networks model, by propagated forward, calculates output
As a result.
(1) in SensorScope wireless sensor network case study on implementation, by No. 2, No. 7, No. 9 destination nodes and neighbours
The space-time characteristic of the real-time stream of node is separately input into S2, S7, S9 model, and calculates output result.
(2) in LWSNDR wireless sensor network case study on implementation, by indoor No. 1 and outdoor No. 4 destination nodes and its neighbour
The real time temperature data flow and real-time humidity data stream for occupying node are separately input into L1-H, L1-T, L2-H, L2-T model, and count
Calculate output result.
Step S5: it is whether abnormal that real-time stream is judged according to model result, and distinguishes failure exception and event anomalies.If
Real-time stream shows that failure or mesh has occurred in the corresponding destination node of real-time stream there are failure exception or event anomalies
Marking near nodal, there are anomalous events, and the abnormality detection and type identification of wireless sensor network are realized with this.
The foregoing is merely presently preferred embodiments of the present invention, all equivalent changes done according to scope of the present invention patent with
Modification, is all covered by the present invention.
Claims (6)
1. a kind of WSN abnormality detection and kind identification method based on data flow space-time characteristic, which is characterized in that including following step
It is rapid:
Step S1: the real time data of wireless sensor network target node and its neighbor node is obtained using space-time sliding window
Stream;
Step S2: the real-time stream that will acquire is mapped to corresponding state space, and is built into the form of Markov chain, is based on
Markov chain calculates state transition probability matrix and crossing condition transition probability matrix extracts the space-time characteristic of real-time stream;
Step S3: building convolutional neural networks models of classifying more, and extraction sample training is more from wireless sensor historical data
Classification convolutional neural networks model, the trained convolutional neural networks models of classifying obtained more;
Step S4: the space-time characteristic of real-time stream is input to trained convolutional neural networks model of classifying more
In, by propagated forward, calculate output result;
Step S5: it is whether abnormal that data flow is judged according to model output result, and distinguishes failure exception and event anomalies.
2. a kind of WSN abnormality detection and kind identification method based on data flow space-time characteristic according to claim 1,
Be characterized in that: the space-time sliding window is made of the detection data at destination node and neighbor node nearest first moment, is led to
Cross the real-time stream that space-time sliding window model obtains destination node and neighbor node.
3. a kind of WSN abnormality detection and kind identification method based on data flow space-time characteristic according to claim 1,
Be characterized in that: the real-time stream that will acquire is mapped to corresponding state space, specifically:
If sensing data sequence is { u1, u2..., ut, utMeasured value for sensor in t moment, utDifference feature be Δ
ut=ut-ut-1;Original series are converted sequence of differences { Δ u by the difference feature of each data first in the sequence of calculation1, Δ
u2..., Δ ut, and sensing data sequence is mapped to corresponding state sky by the size of its difference feature according to 3 σ criterion
Between;State space S includes 9 states altogether, is a, b, c, d, e, f, g, h, k respectively, for any in sensing data sequence
Value utIf utDifference feature Δ utMeet the corresponding conditions in mapping function, then by utBe mapped to state space S=a, b, c,
D, e, f, g, h, k } in corresponding states, mapping function are as follows:
Wherein, σ is sequence of differences { Δ u1, Δ u2..., Δ utStandard deviation, sensing data sequence { u1, u2..., utWarp
It is changed into status switch { s after crossing state mapping1, s2..., st}。
4. a kind of WSN abnormality detection and kind identification method based on data flow space-time characteristic according to claim 3,
It is characterized in that: it is described that the temporal characteristics that state transition probability matrix extracts data flow are calculated based on Markov chain, calculate cross-like
State transition probability matrix extracts the space characteristics of data flow, specific as follows:
(1) calculation method of state transition probability matrix is as follows:
A. the destination node status switch { s after state being mapped1, s2..., stIt is configured to single order Markov chain model X={ X1,
X2..., Xt, the state of t moment is only related with the state at t-1 moment in X model;The state space of X model be S=a, b, c,
D, e, f, g, h, k };
B. the state transition probability p of X model constructed by destination node status switch is calculatedij, pijIndicate model X at the t-1 moment
Present condition s outiAfterwards, go out present condition s in t momentjProbability, wherein si, sj∈S。pijCalculation formula it is as follows:
In formula, N () is used to calculate the total degree of state appearance;
C. all possible state of the destination node status switch in state space S={ a, b, c, d, e, f, g, h, k } is calculated
Transition probability pij, state transition probability matrix P is constituted, the size of matrix P is 9 × 9, matrix are as follows:
Wherein, pij>=0 and
(2) calculation method of crossing condition transition probability matrix is as follows:
A. destination node status switch and neighbor node status switch after state being mapped are built into Markov chain model respectively;
For the neighbor node B of destination node A and node A, the status switch of A, B node is built into Markov chain model respectivelyWithThe state space of model is respectively SA=a, b, c, d, e, f, g, h,
k}、SB={ a, b, c, d, e, f, g, h, k }.
B. the crossing condition transition probability of destination node status switch and neighbor node status switch is then calculated Table
Representation model XAGo out present condition at the t-1 momentAfterwards, model XBGo out present condition in t momentProbability, wherein Calculation formula it is as follows:
Wherein N () is used to calculate the total degree of state appearance;
C. destination node and neighbor node status switch are finally calculated in state space SA={ a, b, c, d, e, f, g, h, k }, SB=
All possible crossing condition transition probability in { a, b, c, d, e, f, g, h, k }Constitute crossing condition transition probability square
Battle array PAB, matrix PABSize be 9 × 9, matrix are as follows:
Wherein,And
5. a kind of WSN abnormality detection and kind identification method based on data flow space-time characteristic according to claim 1,
It is characterized in that: the step S3 specifically:
Step S31: building convolutional neural networks models of classifying, model is by input layer, convolutional layer C1, pond layer S1, convolutional layer more
C2, pond layer S2, full articulamentum FCl, full articulamentum FC2 and output layer are constituted, and totally 8 layers, the activation primitive in model is ReLU
Function;
Step S32: extracting normal, event anomalies, drifting fault and fixed value failure classes sample from sensor historic data,
Composing training sample set and test sample set;
Step S33: being mapped to corresponding state space for training sample, be built into the form of Markov chain, is based on Markov chain
It calculates state transition probability matrix and crossing condition transition probability matrix extracts the space-time characteristic of sample;
Step S34: the space-time characteristic for the training sample being calculated is input to more classification convolutional neural networks models and is instructed
Practice, until model convergence, preservation model structural information and model parameter information.
6. a kind of WSN abnormality detection and kind identification method based on data flow space-time characteristic according to claim 5,
Be characterized in that: the training process of more classification convolutional neural networks models includes that forward and backward is propagated, propagated forward process
In the different characteristic for inputting layer data is extracted by convolutional layer and pond layer, full articulamentum integration characteristics are obtained by Softmax
Classification results, and calculate intersection entropy loss calculate gradient value according to chain rule when back-propagating, pass through stochastic gradient descent method
Update each layer weight.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910518513.4A CN110213788B (en) | 2019-06-15 | 2019-06-15 | WSN (Wireless sensor network) anomaly detection and type identification method based on data flow space-time characteristics |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910518513.4A CN110213788B (en) | 2019-06-15 | 2019-06-15 | WSN (Wireless sensor network) anomaly detection and type identification method based on data flow space-time characteristics |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110213788A true CN110213788A (en) | 2019-09-06 |
CN110213788B CN110213788B (en) | 2021-07-13 |
Family
ID=67792875
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910518513.4A Expired - Fee Related CN110213788B (en) | 2019-06-15 | 2019-06-15 | WSN (Wireless sensor network) anomaly detection and type identification method based on data flow space-time characteristics |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110213788B (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110716843A (en) * | 2019-09-09 | 2020-01-21 | 深圳壹账通智能科技有限公司 | System fault analysis processing method and device, storage medium and electronic equipment |
CN111356108A (en) * | 2020-03-06 | 2020-06-30 | 山东交通学院 | Neural network-based underwater wireless sensor network anomaly diagnosis method |
CN111669373A (en) * | 2020-05-25 | 2020-09-15 | 山东理工大学 | Network anomaly detection method and system based on space-time convolutional network and topology perception |
CN112784896A (en) * | 2021-01-20 | 2021-05-11 | 齐鲁工业大学 | Time series flow data anomaly detection method based on Markov process |
WO2021139251A1 (en) * | 2020-07-30 | 2021-07-15 | 平安科技(深圳)有限公司 | Server system anomaly detection method and apparatus, computer device, and storage medium |
CN113469228A (en) * | 2021-06-18 | 2021-10-01 | 国网山东省电力公司淄博供电公司 | Power load abnormal value identification method based on data flow space-time characteristics |
CN113590654A (en) * | 2021-06-22 | 2021-11-02 | 中国人民解放军国防科技大学 | Spacecraft attitude system anomaly detection method and device based on space-time mode network |
CN113899809A (en) * | 2021-08-20 | 2022-01-07 | 中海石油技术检测有限公司 | In-pipeline detector positioning method based on CNN classification and RNN prediction |
CN113946758A (en) * | 2020-06-30 | 2022-01-18 | 腾讯科技(深圳)有限公司 | Data identification method, device and equipment and readable storage medium |
CN114338853A (en) * | 2021-12-31 | 2022-04-12 | 西南民族大学 | Block chain flow monitoring and detecting method under industrial internet |
CN114781441A (en) * | 2022-04-06 | 2022-07-22 | 电子科技大学 | EEG motor imagery classification method and multi-space convolution neural network model |
WO2023143190A1 (en) * | 2022-01-28 | 2023-08-03 | International Business Machines Corporation | Unsupervised anomaly detection of industrial dynamic systems with contrastive latent density learning |
Citations (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101442807A (en) * | 2008-12-30 | 2009-05-27 | 北京邮电大学 | Method and system for distribution of communication system resource |
US20110246411A1 (en) * | 2010-04-06 | 2011-10-06 | Laneman J Nicholas | Sequence detection methods, devices, and systems for spectrum sensing in dynamic spectrum access networks |
CN102323049A (en) * | 2011-07-18 | 2012-01-18 | 福州大学 | Structural abnormality detection method based on consistent data replacement under incomplete data |
CN102612065A (en) * | 2012-03-19 | 2012-07-25 | 中国地质大学(武汉) | Quick fault-tolerance detection method for monitoring abnormal event by wireless sensor network |
CN102655685A (en) * | 2012-05-29 | 2012-09-05 | 福州大学 | Task fault-tolerance allocation method for wireless sensor networks |
US20140146687A1 (en) * | 2012-11-28 | 2014-05-29 | Huawei Technologies Co., Ltd. | Method and apparatus for remotely locating wireless network fault |
CN105205475A (en) * | 2015-10-20 | 2015-12-30 | 北京工业大学 | Dynamic gesture recognition method |
CN105491614A (en) * | 2016-01-22 | 2016-04-13 | 中国地质大学(武汉) | Wireless sensor network abnormal event detection method and system based on secondary mixed compression |
US20160150438A1 (en) * | 2014-04-04 | 2016-05-26 | Parkervision, Inc. | Momentum transfer communication |
CN105760529A (en) * | 2016-03-03 | 2016-07-13 | 福州大学 | Spatial index and cache construction method for vector data of mobile terminal |
US20170102978A1 (en) * | 2015-10-07 | 2017-04-13 | Business Objects Software Ltd. | Detecting anomalies in an internet of things network |
CN106658590A (en) * | 2016-12-28 | 2017-05-10 | 南京航空航天大学 | Design and implementation of multi-person indoor environment state monitoring system based on WiFi channel state information |
CN106709511A (en) * | 2016-12-08 | 2017-05-24 | 华中师范大学 | Urban rail transit panoramic monitoring video fault detection method based on depth learning |
CN106782504A (en) * | 2016-12-29 | 2017-05-31 | 百度在线网络技术(北京)有限公司 | Audio recognition method and device |
CN106960457A (en) * | 2017-03-02 | 2017-07-18 | 华侨大学 | A kind of colored paintings creative method extracted and scribbled based on image, semantic |
US20180018970A1 (en) * | 2016-07-15 | 2018-01-18 | Google Inc. | Neural network for recognition of signals in multiple sensory domains |
CN109447263A (en) * | 2018-11-07 | 2019-03-08 | 任元 | A kind of space flight accident detection method based on generation confrontation network |
CN109640335A (en) * | 2019-02-28 | 2019-04-16 | 福建师范大学 | Wireless sensor fault diagnosis algorithm based on convolutional neural networks |
-
2019
- 2019-06-15 CN CN201910518513.4A patent/CN110213788B/en not_active Expired - Fee Related
Patent Citations (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101442807A (en) * | 2008-12-30 | 2009-05-27 | 北京邮电大学 | Method and system for distribution of communication system resource |
US20110246411A1 (en) * | 2010-04-06 | 2011-10-06 | Laneman J Nicholas | Sequence detection methods, devices, and systems for spectrum sensing in dynamic spectrum access networks |
CN102323049A (en) * | 2011-07-18 | 2012-01-18 | 福州大学 | Structural abnormality detection method based on consistent data replacement under incomplete data |
CN102612065A (en) * | 2012-03-19 | 2012-07-25 | 中国地质大学(武汉) | Quick fault-tolerance detection method for monitoring abnormal event by wireless sensor network |
CN102655685A (en) * | 2012-05-29 | 2012-09-05 | 福州大学 | Task fault-tolerance allocation method for wireless sensor networks |
US20140146687A1 (en) * | 2012-11-28 | 2014-05-29 | Huawei Technologies Co., Ltd. | Method and apparatus for remotely locating wireless network fault |
US20160150438A1 (en) * | 2014-04-04 | 2016-05-26 | Parkervision, Inc. | Momentum transfer communication |
US20170102978A1 (en) * | 2015-10-07 | 2017-04-13 | Business Objects Software Ltd. | Detecting anomalies in an internet of things network |
CN105205475A (en) * | 2015-10-20 | 2015-12-30 | 北京工业大学 | Dynamic gesture recognition method |
CN105491614A (en) * | 2016-01-22 | 2016-04-13 | 中国地质大学(武汉) | Wireless sensor network abnormal event detection method and system based on secondary mixed compression |
CN105760529A (en) * | 2016-03-03 | 2016-07-13 | 福州大学 | Spatial index and cache construction method for vector data of mobile terminal |
US20180018970A1 (en) * | 2016-07-15 | 2018-01-18 | Google Inc. | Neural network for recognition of signals in multiple sensory domains |
CN106709511A (en) * | 2016-12-08 | 2017-05-24 | 华中师范大学 | Urban rail transit panoramic monitoring video fault detection method based on depth learning |
CN106658590A (en) * | 2016-12-28 | 2017-05-10 | 南京航空航天大学 | Design and implementation of multi-person indoor environment state monitoring system based on WiFi channel state information |
CN106782504A (en) * | 2016-12-29 | 2017-05-31 | 百度在线网络技术(北京)有限公司 | Audio recognition method and device |
CN106960457A (en) * | 2017-03-02 | 2017-07-18 | 华侨大学 | A kind of colored paintings creative method extracted and scribbled based on image, semantic |
CN109447263A (en) * | 2018-11-07 | 2019-03-08 | 任元 | A kind of space flight accident detection method based on generation confrontation network |
CN109640335A (en) * | 2019-02-28 | 2019-04-16 | 福建师范大学 | Wireless sensor fault diagnosis algorithm based on convolutional neural networks |
Non-Patent Citations (1)
Title |
---|
刘莘: "基于时空分析的CCS泄漏预警关键技术研究", 《工程科技Ⅰ辑》 * |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110716843B (en) * | 2019-09-09 | 2022-11-22 | 深圳壹账通智能科技有限公司 | System fault analysis processing method and device, storage medium and electronic equipment |
CN110716843A (en) * | 2019-09-09 | 2020-01-21 | 深圳壹账通智能科技有限公司 | System fault analysis processing method and device, storage medium and electronic equipment |
CN111356108A (en) * | 2020-03-06 | 2020-06-30 | 山东交通学院 | Neural network-based underwater wireless sensor network anomaly diagnosis method |
CN111356108B (en) * | 2020-03-06 | 2022-04-19 | 山东交通学院 | Neural network-based underwater wireless sensor network anomaly diagnosis method |
CN111669373A (en) * | 2020-05-25 | 2020-09-15 | 山东理工大学 | Network anomaly detection method and system based on space-time convolutional network and topology perception |
CN111669373B (en) * | 2020-05-25 | 2022-04-01 | 山东理工大学 | Network anomaly detection method and system based on space-time convolutional network and topology perception |
CN113946758A (en) * | 2020-06-30 | 2022-01-18 | 腾讯科技(深圳)有限公司 | Data identification method, device and equipment and readable storage medium |
WO2021139251A1 (en) * | 2020-07-30 | 2021-07-15 | 平安科技(深圳)有限公司 | Server system anomaly detection method and apparatus, computer device, and storage medium |
CN112784896A (en) * | 2021-01-20 | 2021-05-11 | 齐鲁工业大学 | Time series flow data anomaly detection method based on Markov process |
CN113469228A (en) * | 2021-06-18 | 2021-10-01 | 国网山东省电力公司淄博供电公司 | Power load abnormal value identification method based on data flow space-time characteristics |
CN113590654A (en) * | 2021-06-22 | 2021-11-02 | 中国人民解放军国防科技大学 | Spacecraft attitude system anomaly detection method and device based on space-time mode network |
CN113899809A (en) * | 2021-08-20 | 2022-01-07 | 中海石油技术检测有限公司 | In-pipeline detector positioning method based on CNN classification and RNN prediction |
CN113899809B (en) * | 2021-08-20 | 2024-02-27 | 中海石油技术检测有限公司 | In-pipeline detector positioning method based on CNN classification and RNN prediction |
CN114338853A (en) * | 2021-12-31 | 2022-04-12 | 西南民族大学 | Block chain flow monitoring and detecting method under industrial internet |
WO2023143190A1 (en) * | 2022-01-28 | 2023-08-03 | International Business Machines Corporation | Unsupervised anomaly detection of industrial dynamic systems with contrastive latent density learning |
CN114781441A (en) * | 2022-04-06 | 2022-07-22 | 电子科技大学 | EEG motor imagery classification method and multi-space convolution neural network model |
CN114781441B (en) * | 2022-04-06 | 2024-01-26 | 电子科技大学 | EEG motor imagery classification method and multi-space convolution neural network model |
Also Published As
Publication number | Publication date |
---|---|
CN110213788B (en) | 2021-07-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110213788A (en) | WSN abnormality detection and kind identification method based on data flow space-time characteristic | |
CN112241814B (en) | Traffic prediction method based on reinforced space-time diagram neural network | |
CN111209434B (en) | Substation equipment inspection system and method based on multi-source heterogeneous data fusion | |
CN110135319A (en) | A kind of anomaly detection method and its system | |
CN109063728A (en) | A kind of fire image deep learning mode identification method | |
CN111814661B (en) | Human body behavior recognition method based on residual error-circulating neural network | |
CN113240688A (en) | Integrated flood disaster accurate monitoring and early warning method | |
CN109829893A (en) | A kind of defect object detection method based on attention mechanism | |
CN104992223A (en) | Intensive population estimation method based on deep learning | |
CN103268495B (en) | Human body behavior modeling recognition methods based on priori knowledge cluster in computer system | |
CN107818302A (en) | Non-rigid multi-scale object detection method based on convolutional neural network | |
CN107341452A (en) | Human bodys' response method based on quaternary number space-time convolutional neural networks | |
CN109800736A (en) | A kind of method for extracting roads based on remote sensing image and deep learning | |
CN107301387A (en) | A kind of image Dense crowd method of counting based on deep learning | |
CN106991666B (en) | A kind of disease geo-radar image recognition methods suitable for more size pictorial informations | |
CN109214308A (en) | A kind of traffic abnormity image identification method based on focal loss function | |
CN108596203A (en) | Optimization method of the pond layer in parallel to pantograph carbon slide surface abrasion detection model | |
CN111046961B (en) | Fault classification method based on bidirectional long-time and short-time memory unit and capsule network | |
CN108009477A (en) | Stream of people's quantity detection method, device, storage medium and the electronic equipment of image | |
CN114399719B (en) | Transformer substation fire video monitoring method | |
CN110533103A (en) | A kind of lightweight wisp object detection method and system | |
CN109800712A (en) | A kind of vehicle detection method of counting and equipment based on depth convolutional neural networks | |
CN113361686A (en) | Multilayer heterogeneous multi-mode convolutional neural network integrated robot inspection method | |
CN115629160A (en) | Air pollutant concentration prediction method and system based on space-time diagram | |
CN115544239A (en) | Deep learning model-based layout preference prediction method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
CB03 | Change of inventor or designer information | ||
CB03 | Change of inventor or designer information |
Inventor after: Wu Qunyong Inventor after: Deng Li Inventor before: Wu Qunyong Inventor before: Deng Liping |
|
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210713 |