CN109525598A - A kind of fault-tolerant compression method of wireless sense network depth and system based on variation mixing - Google Patents
A kind of fault-tolerant compression method of wireless sense network depth and system based on variation mixing Download PDFInfo
- Publication number
- CN109525598A CN109525598A CN201811603750.2A CN201811603750A CN109525598A CN 109525598 A CN109525598 A CN 109525598A CN 201811603750 A CN201811603750 A CN 201811603750A CN 109525598 A CN109525598 A CN 109525598A
- Authority
- CN
- China
- Prior art keywords
- network
- training
- variation
- compression
- data
- 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.)
- Pending
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L69/00—Network arrangements, protocols or services independent of the application payload and not provided for in the other groups of this subclass
- H04L69/04—Protocols for data compression, e.g. ROHC
-
- 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
-
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W52/00—Power management, e.g. TPC [Transmission Power Control], power saving or power classes
- H04W52/02—Power saving arrangements
- H04W52/0209—Power saving arrangements in terminal devices
- H04W52/0212—Power saving arrangements in terminal devices managed by the network, e.g. network or access point is master and terminal is slave
- H04W52/0222—Power saving arrangements in terminal devices managed by the network, e.g. network or access point is master and terminal is slave in packet switched networks
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Abstract
It include: to be denoised to raw temperature data, go abnormality processing, and be normalized the present invention provides a kind of fault-tolerant compression method of wireless sense network depth based on variation mixing and system, method；RBM is generated into model and variation hybrid coder combined structure goes out variation combined errors compression network；Using processed sensing data collection as the training set of network, it is iterated training, improves data compression rate and reconstruction accuracy.The beneficial effects of the present invention are: technical solution proposed by the present invention effectively reduces the node energy consumption consumption of wireless sense network, and there is good compression ratio and robustness, the life duty cycle of wireless sense network is improved, proposes a kind of new method to improve WSN timeliness, energy saving and reliability.
Description
Technical field
The present invention relates to wireless sense network and depth learning technology field more particularly to it is a kind of based on variation mixing it is wireless
The fault-tolerant compression method of Sensor Network depth and system.
Background technique
With the continuous development of computer and the relevant technologies, so that sensing, communication and the functions such as calculating are integrated in one
Small module is possibly realized, to promote the deep development and extensive use of wireless sensor network (WSN) technology.People can be with
Objective world is perceived by wireless sensor network, without going the area of observation coverage to be observed in person.But due to wireless sensor network
Particularity, development still face many problems to be solved.
Since node computing capability, storage capacity, communication bandwidth and power supply energy are all very limited, while again due to wirelessly passing
Feeling the initial data that great deal of nodes obtains in network, there are serious data redundancies, including same the acquired number of node adjacent moment
Temporal redundancy caused by similitude between, adjacent node is in mutually sky caused by the similitude between acquired data in the same time
Domain redundancy.If directly transmitting these initial data, not only limited bandwidth resources can be made to be not fully utilized, while can be due to
The original data volume to be transmitted is too big and system is caused to exist compared with long time delay, serious to affect entire wireless sensing network system
To the real-time monitoring of monitoring object, and there are the initial data of bulk redundancy can exhaust the limited energy of node quickly for transmission,
Seriously affect the service life of whole system.For data redundancy present in wireless sensor network, guaranteeing the small feelings of loss of significance
Under condition, seeming to wireless sensor network progress data compression is highly desirable.
The existing data compression algorithm for wireless sensor network according to different classification standards can be divided into it is lossless with
Damage algorithm, distributed and local algorithm, shallow-layer compression and depth-compression algorithm etc..It is calculated for shallow-layer compression with depth-compression
The WSN data stream compression algorithm based on temporal correlation is summarized by method, domestic scholars Lee China, state, and India scientist Reshma proposes fortune
Row length coding is applicable in the compression of WSN flow data very much.Scholar Ying Bei China proposes to be based on space phase on international conference ICACT in 2016
The data lightweight compression algorithm of closing property saves energy more more than wavelet compression in the case where identical distortion rate.It is above-mentioned
It is shallow-layer compression algorithm, it is weaker to the learning ability of depth high-order feature.U.S. academician professor Hinton proposes depth nerve
Network data compression method, the theory primarily point out: deep-neural-network RBM can learn the nonlinear characteristic of high dimension vector,
And feature visualization may be implemented.The layer-by-layer pre-training of network preferably solves the gradient of deep neural network model training more
The problem of dissipating.Global more excellent weight parameter can be obtained and convergence depth is fast by being adjusted by contrast divergence algorithm.
The research that deep learning model is applied to WSN data compression at present is less, and only Qiu Li is based on stack up to proposition
Data compression is carried out from encoding model, this method combines storehouse autocoder and WSN cluster Routing Protocol, presses with tradition
Compression algorithm, which is compared, to improve 7.5 percentage points for data fusion precision.It can be by deep learning model in conclusion finding one kind
Algorithm applied to WSN data compression is necessary.
Summary of the invention
To solve the above-mentioned problems, the present invention provides a kind of fault-tolerant compressions of wireless sense network depth based on variation mixing
A kind of method and system, fault-tolerant compression method of wireless sense network depth based on variation mixing, mainly comprise the steps that
S101: the temperature data of each terminal sensing node of wireless sense network, and the temperature number that will be obtained from each node are obtained
According to the training dataset of composition corresponding node；
S102: the temperature data concentrated to the training data pre-processes, and obtains pretreated training dataset；
S103: building stack RBM network carries out pre-training to stack RBM network using pretreated training dataset,
The weight of stack RBM network model after obtaining pre-training and biasing；
S104: by the stack RBM network model and variation hybrid coder models coupling after pre-training, it is mixed to construct variation
Close mistake compression network；
S105: variation combined errors compression network is trained using the pretreated training dataset, is obtained
Variation combined errors compression network after training；
S106: obtaining the true temperature data of each terminal sensing node of wireless sense network, true by what is obtained from corresponding node
The test data set of real temperature data composition corresponding node；And the variation combined errors after training test data set input
Compression network obtains the compressed data of corresponding node true temperature data.
Further, in step S101, the temperature data is that Intel, University of California laboratory wireless sensor network is ground
Study carefully team, places the ambient temperature data letter that 54 sensor nodes are collected in laboratory from 2 28th, 2004 to April 5
It ceases, the timestamp topology information of collection in sensor node every 31 seconds, totally 3308442 temperature datas.
Further, in step S102, pretreated to the temperature data progress of training data concentration steps are as follows:
S201: removing and be less than -5 DEG C and the abnormal temperature data greater than 45 DEG C in the training dataset, and it is pre- to obtain first
Handle data set；
S202: removing other abnormal datas that first preprocessed data is concentrated using triple standard difference method, obtains the
Two preprocessed data collection；
S203: by the second preprocessed data collection normalized mapping to [0,1] section, pretreated trained number is obtained
According to collection.
Further, in step S203, method for normalizing uses minimax method for normalizing.
Further, it in step S103, is stacked gradually using four RBM networks and constitutes stack RBM network；Last RBM net
Visual layers of the hidden layer of network as next RBM network successively extract feature, successively dimensionality reduction；And utilize the layer-by-layer data compression of RBM
Pre-training algorithm carries out pre-training to stack RBM network.
Further, in step S104, the stack RBM coding network that training obtains is made into symmetrical transposition, obtains stack RBM
Decoding network；Using stack RBM encoding and decoding network as the codec of VAE, and two groups of data of VAE encoder output are distinguished
As the mean value and variance of Gaussian Profile, the input of VAE decoder is generated by the method that variation samples, and it is poor to constitute variation mixing
Wrong compression network.
Further, in step S105, using the pretreated training dataset to variation combined errors compressed web
The method that network is trained are as follows: initialize variation combined errors pressure using the stack RBM encoding and decoding network parameter after training first
The parameter of contracting network, the variation combined errors compression network after being initialized；
Then the variation combined errors compression network after initialization is carried out using the pretreated training dataset
Training, the variation combined errors compression network after being trained；Training method are as follows: compressed using variation combined errors and reconstruct calculation
Method carries out small parameter perturbations to the variation combined errors compression network after initialization.
Further, in step S106, the method for the true temperature data of acquisition each terminal sensing node of wireless sense network
Are as follows: each terminal sensing node temperature collection flow data of wireless sense network is acquired using sensor.
Further, a kind of fault-tolerant compressibility of wireless sense network depth based on variation mixing, it is characterised in that: including
Following steps:
Data acquisition module, for obtaining the temperature data of each terminal sensing node of wireless sense network, and will be from each node
The training dataset of the temperature data composition corresponding node of acquisition；
Pre-training module, the temperature data for concentrating to the training data are pre-processed, are obtained pretreated
Training dataset；
Model buildings module, for constructing stack RBM network, using pretreated training dataset to stack RBM net
Network carries out pre-training, the weight of the stack RBM network model after obtaining pre-training and biasing；
Net structure module, for by the stack RBM network model and variation hybrid coder models coupling after pre-training,
Construct variation combined errors compression network；
Network training module, for using the pretreated training dataset to variation combined errors compression network into
Row training, the variation combined errors compression network after being trained；
Test module will be from correspondence for obtaining the true temperature data for obtaining each terminal sensing node of wireless sense network
Node obtain true temperature data group at corresponding node test data set；And it will be after test data set input training
Variation combined errors compression network obtains the compressed data for corresponding to the node true temperature data.
Further, in pre-training module, pretreated step packet is carried out to the temperature data that the training data is concentrated
It includes with lower unit:
Abnormity removing unit, for removing the abnormal temperature number in the training dataset less than -5 DEG C and greater than 45 DEG C
According to obtaining the first preprocessed data collection；
Standard deviation units, for removing other abnormal numbers that first preprocessed data is concentrated using triple standard difference method
According to obtaining the second preprocessed data collection；
Normalization unit, for [0,1] section, being pre-processed the second preprocessed data collection normalized mapping
Training dataset afterwards.
Technical solution provided by the invention has the benefit that technical solution proposed by the present invention effectively reduces
The node energy consumption of wireless sense network consumes, and has good compression ratio and robustness, improves the life work of wireless sense network
Make the period, proposes a kind of new method to improve WSN timeliness, energy saving and reliability.
Detailed description of the invention
Present invention will be further explained below with reference to the attached drawings and examples, in attached drawing:
Fig. 1 is a kind of process of the fault-tolerant compression method of wireless sense network depth based on variation mixing in the embodiment of the present invention
Figure；
Fig. 2 (a) is original temperature sequence data sequence in the embodiment of the present invention；
Fig. 2 (b) is that temperature sequence after threshold value is set in the embodiment of the present invention；
Fig. 2 (c) is that the temperature data sequence after three times standard deviation is carried out in the embodiment of the present invention；
Fig. 2 (d) is temperature data sequence after normalizing in the embodiment of the present invention；
Fig. 3 is stack RBM encoding and decoding network architecture schematic diagram in the embodiment of the present invention；
Fig. 4 is variation combined errors compression network structural schematic diagram in the embodiment of the present invention；
Fig. 5 is a kind of module of the fault-tolerant compressibility of wireless sense network depth based on variation mixing in the embodiment of the present invention
Composition schematic diagram；
Fig. 6 be in the embodiment of the present invention variation combined errors compression network to the reconstruction result schematic diagram of training sample.
Specific embodiment
For a clearer understanding of the technical characteristics, objects and effects of the present invention, now control attached drawing is described in detail
A specific embodiment of the invention.
The embodiment provides a kind of fault-tolerant compression method of wireless sense network depth based on variation mixing and it is
System.
Referring to FIG. 1, Fig. 1 is a kind of fault-tolerant compression of wireless sense network depth based on variation mixing in the embodiment of the present invention
The flow chart of method, specifically comprises the following steps:
S101: the temperature data of each terminal sensing node of wireless sense network, and the temperature number that will be obtained from each node are obtained
According to the training dataset of composition corresponding node；
S102: the temperature data concentrated to the training data pre-processes, and obtains pretreated training dataset；
S103: building stack RBM network carries out pre-training to stack RBM network using pretreated training dataset,
The weight of stack RBM network model after obtaining pre-training and biasing；
S104: by the stack RBM network model and variation hybrid coder models coupling after pre-training, it is mixed to construct variation
Close mistake compression network；
The method of stack RBM network model and variation hybrid coder models coupling after pre-training is as follows:
First using the stack RBM network model after pre-training as the encoder of variation hybrid coder (VAE) model, so
The stack RBM network model after pre-training is subjected to transposition afterwards, and using the RBM network model after transposition as variation hybrid decoding
The encoder of device (VAE) model, and then construct variation combined errors compression network；
S105: variation combined errors compression network is trained using the pretreated training dataset, is obtained
Variation combined errors compression network after training；
S106: obtaining the true temperature data of each terminal sensing node of wireless sense network, true by what is obtained from corresponding node
The test data set of real temperature data composition corresponding node；And the variation combined errors after training test data set input
Compression network obtains the compressed data of corresponding node true temperature data.
In step S101, the temperature data is Intel, University of California laboratory wireless sensor network research team, from
The ambient temperature data information that 54 sensor nodes are collected, sensor were placed in laboratory to April 5 on 2 28th, 2004
The timestamp topology information of collection in node every 31 seconds, totally 3308442 temperature datas.
In step S102, pretreated to the temperature data progress of training data concentration steps are as follows:
S201: removing and be less than -5 DEG C and the abnormal temperature data greater than 45 DEG C in the training dataset, and it is pre- to obtain first
Handle data set；
S202: removing other abnormal datas that first preprocessed data is concentrated using triple standard difference method, obtains the
Two preprocessed data collection；
S203: by the second preprocessed data collection normalized mapping to [0,1] section, pretreated trained number is obtained
According to collection.
After temperature of the above-mentioned steps for each node pre-processes, with the temperature of Intel's laboratory interior joint 7
Degree is according to the training set as model, and temperature sequence is as shown in Fig. 2, wherein Fig. 2 (a) is original temperature sequence data sequence, Fig. 2
(b) to set temperature sequence (the first preprocessed data collection) after threshold value, Fig. 2 (c) is the temperature data sequence carried out after three times standard deviation
It arranges (the second preprocessed data collection), Fig. 2 (d) is temperature data sequence (pretreated training dataset) after normalization.
In step S203, method for normalizing uses minimax method for normalizing.
In step S103, is stacked gradually using four RBM networks and constitute stack RBM network；The hidden layer of last RBM network
As the visual layers of next RBM network, feature, successively dimensionality reduction are successively extracted；And utilize the layer-by-layer data compression pre-training algorithm of RBM
Pre-training is carried out to the stack RBM network.
In step S104, the stack RBM coding network that training obtains is made into symmetrical transposition, obtains stack RBM decoding network
(stack RBM encoding and decoding network architecture is as shown in Figure 3)；Using stack RBM encoding and decoding network as the codec of VAE, and
Using two groups of data of VAE encoder output as the mean value of Gaussian Profile and variance, generated by the method that variation samples
The input of VAE decoder is constituted variation combined errors compression network (as described in Figure 4).
In step S105, variation combined errors compression network is trained using the pretreated training dataset
Method are as follows: first using training after stack RBM encoding and decoding network parameter initialization variation combined errors compression network ginseng
Number, the variation combined errors compression network after being initialized；
Then the variation combined errors compression network after initialization is carried out using the pretreated training dataset
Training, the variation combined errors compression network after being trained: is compressed using variation combined errors and restructing algorithm is to initialization
Variation combined errors compression network afterwards carries out small parameter perturbations.
In step S106, the method for the true temperature data of acquisition each terminal sensing node of wireless sense network are as follows: utilize biography
Sensor is acquired each terminal sensing node temperature collection flow data of wireless sense network.
As shown in figure 5, for a kind of fault-tolerant compression system of wireless sense network depth based on variation mixing in the embodiment of the present invention
The module composition schematic diagram of system, it is characterised in that: including sequentially connected data acquisition module 11, pre-training module 12, model
Build module 13, net structure module 14, network training module 15 and test module 16；
Data acquisition module 11, for obtaining the temperature data of each terminal sensing node of wireless sense network, and will be from each section
The training dataset for the temperature data composition corresponding node that point obtains；
Pre-training module 12, the temperature data for concentrating to the training data pre-processes, after obtaining pretreatment
Training dataset；
Model buildings module 13, for constructing stack RBM network, using pretreated training dataset to stack RBM
Network carries out pre-training, the weight of the stack RBM network model after obtaining pre-training and biasing；
Net structure module 14, for by the stack RBM network model and variation hybrid coder model knot after pre-training
It closes, constructs variation combined errors compression network；
Network training module 15, for using the pretreated training dataset to variation combined errors compression network
It is trained, the variation combined errors compression network after being trained；
Test module 16 will be saved for obtaining the true temperature data of each terminal sensing node of wireless sense network from corresponding
The true temperature data group of acquisition is put into the test data set of corresponding node；And the change after training test data set input
Divide combined errors compression network, obtains the compressed data for corresponding to the node true temperature data.
In the present embodiment, in pre-training module 12, the temperature data that the training data is concentrated is carried out pretreated
Step includes with lower unit:
Abnormity removing unit, for removing the abnormal temperature number in the training dataset less than -5 DEG C and greater than 45 DEG C
According to obtaining the first preprocessed data collection；
Standard deviation units, for removing other abnormal numbers that first preprocessed data is concentrated using triple standard difference method
According to obtaining the second preprocessed data collection；
Normalization unit, for [0,1] section, being pre-processed the second preprocessed data collection normalized mapping
Training dataset afterwards.
It is layer-by-layer by RBM using pretreated training dataset as the training set of stack RBM network in the present embodiment
Data compression pre-training algorithm carries out pre-training to stack RBM network, and algorithm flow is as follows:
Step:
1. the parameter of pair stack RBM encoding and decoding network initializes, Δ W=0, Δ b=0, Δ c=0；
2. setting the number of iterations initializaing variable iter=0；
3. carrying out propagated forward for each of training set training sample, obtaining all hidden layer unit activatings
Value；
4. pair all hiding layer units seek corresponding visual layers unit activating value；
5. seeking visual layers each unit precision, local derviation numerical value is calculated；
6. W is updated,
7.iter=iter+1；
8. if exporting step 6 result iter < maxiter, returns to step 3, be otherwise finished
After the completion of variation combined errors compression network model is by the training of above-mentioned algorithm, in the compression and reconstruct of temperature data
On it is with good performance.Technical solution proposed by the invention will be done further by several description of test below
It illustrates:
Experiment 1: reconstruction result of the variation combined errors compression network to training sample
Stack RBM coding/decoding module the number of iterations is 2000, BP back propagation learning in variation combined errors compression network
Number is 500 times.Entire model initial vector dimension 120, compression multiple are 10 times, and training data takes in Intel laboratory and saves
The temperature data of point 7.The data point that box marks in Fig. 6 is the raw temperature data in training set, the data of solid dot mark
Point is the temperature data of variation combined errors compression network reconstruct.
From fig. 6 it can be seen that variation combined errors compression network compression multiple be 10 times in the case where, reconstruction accuracy
It is higher, it can be better close to raw temperature data.
Experiment 2: data reconstruction error of the same node under different compression ratios
Stack RBM coding/decoding module the number of iterations is 2000, BP back propagation learning in variation combined errors compression network
Number is 500 times.Training data is the temperature data of 1 to No. 10 node, and entire model initial vector dimension 120 passes through adjusting
Layer 6 neuronal quantity changes the compression multiple of model, and compression multiple is successively adjusted to 24,8,4,2,1 times.
Experimental result is as shown in table 1:
Reconstructed error of the same node of table 1 under different compression ratios
As shown in Table 1: on a single node, with the increase of hiding intrinsic dimensionality, the i.e. reduction of compression multiple, node
Reconstructed error increases or reduces there is no apparent, and what this illustrated that deep learning algorithm obtains is data build-in attribute, intrinsic special
Sign shows on weight matrix and unrelated with compressed dimension that this algorithm has very strong robustness, this is deep layer compression side
The difference of method and shallow-layer compression method, shallow-layer compression method often with compression multiple increase reconstructed error can occur it is bright
It is aobvious to increase, while also embodying the correlation of same node temperature sequence in time.Under identical hidden layer intrinsic dimensionality, i.e.,
When identical compression multiple, the reconstructed error approximation of different nodes is convergent, this illustrates the versatility of deep learning model, similarly
Model can learn the fluctuation of the time series to different nodes.
Experiment 3: the model transfer learning experiment of this algorithm on different nodes
To verify the spatial coherence between the Generalization Capability of variation combined errors compression network and different nodes, first with section
The data of point 7 do training, are tested under the network parameter that the node 1~20 in wireless sense network is trained herein, this is in depth
It is referred to as transfer learning in degree study.Meanwhile by the test error of each node obtained above with individually to each node into
Row training, i.e., each node have the weighting parameter of training pattern corresponding to this node flow data, and two experiments are obtained
The reconstructed error of each node as a comparison, the ability of the model transfer learning of verification algorithm on different nodes.Hidden layer is special
Levying dimension is 5 dimensions, i.e., compression multiple is 24 times, RBM pre-training 2000 times, BP backpropagation 200 times.Experimental result such as table 2
It is shown:
The model transfer learning (1~node of node 20) of table 2 on different nodes
It can be seen that test error and each node oneself train obtained training error very close from table 2, explanation
Data space correlation in same WSN at each node is high, also demonstrate variation combined errors compression network have it is certain general
Change ability
In conclusion the invention proposes a kind of fault-tolerant compression method of wireless sense network depth based on variation mixing, it should
The stack RBM encoding and decoding network that method is stacked into using processed four RBM of sensing data collection training first, utilizes Ma Erke
Husband's random field property, non-directed graph decomposing property and gibbs sampler and K step contrast divergence algorithm are trained each RBM.
After the completion of training, the stack RBM coding network parameter that training obtains is made into simple symmetrical transposition, stack RBM decoding can be obtained
Network, and then obtain stack RBM encoding and decoding network model.Using stack RBM encoding and decoding network as the codec of VAE, respectively
Using two groups of data of VAE encoder output as the mean value of Gaussian Profile and variance, is sampled by variation and generate VAE decoder
Input constitutes variation combined errors compression network.Retraining is carried out to variation combined errors compression network using training dataset,
Improve data compression rate and reconstruction accuracy.
The beneficial effects of the present invention are: technical solution proposed by the present invention effectively reduces the node energy of wireless sense network
Consumption consumption, and have good compression ratio and robustness, improve the life duty cycle of wireless sense network, for improve WSN and
Shi Xing, energy saving and reliability propose a kind of new method.
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and
Within principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (10)
1. a kind of fault-tolerant compression method of wireless sense network depth based on variation mixing, it is characterised in that: the following steps are included:
S101: the temperature data of each terminal sensing node of wireless sense network, and the temperature data group that will be obtained from each node are obtained
At the training dataset of corresponding node；
S102: the temperature data concentrated to the training data pre-processes, and obtains pretreated training dataset；
S103: building stack RBM network carries out pre-training to stack RBM network using pretreated training dataset, obtains
The weight of stack RBM network model after pre-training and biasing；
S104: by the stack RBM network model and variation hybrid coder VAE models coupling after pre-training, it is mixed to construct variation
Close mistake compression network；
S105: variation combined errors compression network is trained using the pretreated training dataset, is trained
Variation combined errors compression network afterwards；
S106: the true temperature data of each terminal sensing node of wireless sense network, the true temperature that will be obtained from corresponding node are obtained
Degree is according to the test data set for forming corresponding node；And the variation combined errors after test data set input training are compressed
Network obtains the compressed data of corresponding node true temperature data.
2. a kind of fault-tolerant compression method of wireless sense network depth based on variation mixing as described in claim 1, feature exist
In: in step S101, the temperature data was Intel, University of California laboratory wireless sensor network research team, from 2004
The ambient temperature data information that 54 sensor nodes are collected was placed in laboratory to April 5 within 28 days 2 months, sensor node is every
The timestamp topology information of collection in 31 seconds, totally 3308442 temperature datas.
3. a kind of fault-tolerant compression method of wireless sense network depth based on variation mixing as described in claim 1, feature exist
In: in step S102, pretreated to the temperature data progress of training data concentration steps are as follows:
S201: it removes and is less than -5 DEG C and the abnormal temperature data greater than 45 DEG C in the training dataset, obtain the first pretreatment
Data set；
S202: other abnormal datas that first preprocessed data is concentrated are removed using triple standard difference method, it is pre- to obtain second
Handle data set；
S203: by the second preprocessed data collection normalized mapping to [0,1] section, pretreated training data is obtained
Collection.
4. a kind of fault-tolerant compression method of wireless sense network depth based on variation mixing as claimed in claim 3, feature exist
In: in step S203, method for normalizing uses minimax method for normalizing.
5. a kind of fault-tolerant compression method of wireless sense network depth based on variation mixing as described in claim 1, feature exist
In: in step S103, is stacked gradually using four RBM networks and constitute stack RBM network；The hidden layer conduct of last RBM network
The visual layers of next RBM network successively extract feature, successively dimensionality reduction；And using the layer-by-layer data compression pre-training algorithm of RBM to stack
Formula RBM network carries out pre-training.
6. a kind of fault-tolerant compression method of wireless sense network depth based on variation mixing as described in claim 1, feature exist
In: in step S104, the stack RBM coding network that training obtains is made into symmetrical transposition, obtains stack RBM decoding network；By stack
Codec of the formula RBM encoding and decoding network as VAE, and using two groups of data of VAE encoder output as Gaussian Profile
Mean value and variance, pass through variation sample method generate VAE decoder input, constitute variation combined errors compression network.
7. a kind of fault-tolerant compression method of wireless sense network depth based on variation mixing as described in claim 1, feature exist
In: in step S105, variation combined errors compression network is trained using the pretreated training dataset side
Method are as follows: first using the parameter of the stack RBM encoding and decoding network parameter initialization variation combined errors compression network after training, obtain
Variation combined errors compression network after to initialization；
Then the variation combined errors compression network after initialization is trained using the pretreated training dataset,
Variation combined errors compression network after being trained；Training method are as follows: use the compression of variation combined errors and restructing algorithm pair
Variation combined errors compression network after initialization carries out small parameter perturbations.
8. a kind of fault-tolerant compression method of wireless sense network depth based on variation mixing as described in claim 1, feature exist
In: in step S106, the method for the true temperature data of acquisition each terminal sensing node of wireless sense network are as follows: utilize sensor pair
The each terminal sensing node temperature collection flow data of wireless sense network is acquired.
9. a kind of fault-tolerant compressibility of wireless sense network depth based on variation mixing, it is characterised in that: the following steps are included:
Data acquisition module for obtaining the temperature data of each terminal sensing node of wireless sense network, and will be obtained from each node
Temperature data composition corresponding node training dataset；
Pre-training module, the temperature data for concentrating to the training data pre-process, and obtain pretreated training
Data set；
Model buildings module, for constructing stack RBM network, using pretreated training dataset to stack RBM network into
Row pre-training, the weight of the stack RBM network model after obtaining pre-training and biasing；
Net structure module, for constructing the stack RBM network model and variation hybrid coder models coupling after pre-training
Variation combined errors compression network out；
Network training module, for being instructed using the pretreated training dataset to variation combined errors compression network
Practice, the variation combined errors compression network after being trained；
Test module, for obtaining the true temperature data of each terminal sensing node of wireless sense network, by what is obtained from each node
True temperature data group at corresponding node test data set；And it is the variation mixing after test data set input training is poor
Wrong compression network obtains the compressed data of corresponding node true temperature data.
10. a kind of fault-tolerant compressibility of wireless sense network depth based on variation mixing as claimed in claim 9, feature exist
In: in pre-training module, carrying out pretreated step to the temperature data that the training data is concentrated includes with lower unit:
Abnormity removing unit is obtained for removing the abnormal temperature data in the training dataset less than -5 DEG C and greater than 45 DEG C
To the first preprocessed data collection；
Standard deviation units, for removing other abnormal datas that first preprocessed data is concentrated using triple standard difference method,
Obtain the second preprocessed data collection；
Normalization unit, for [0,1] section, obtaining pretreated the second preprocessed data collection normalized mapping
Training dataset.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811603750.2A CN109525598A (en) | 2018-12-26 | 2018-12-26 | A kind of fault-tolerant compression method of wireless sense network depth and system based on variation mixing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811603750.2A CN109525598A (en) | 2018-12-26 | 2018-12-26 | A kind of fault-tolerant compression method of wireless sense network depth and system based on variation mixing |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109525598A true CN109525598A (en) | 2019-03-26 |
Family
ID=65797145
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811603750.2A Pending CN109525598A (en) | 2018-12-26 | 2018-12-26 | A kind of fault-tolerant compression method of wireless sense network depth and system based on variation mixing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109525598A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110446173A (en) * | 2019-07-31 | 2019-11-12 | 中国地质大学（武汉） | A kind of energy-efficient satellite-carried wireless Sensor Network data compression method |
| CN112153659A (en) * | 2020-08-21 | 2020-12-29 | 中国地质大学（武汉） | Efficient and energy-saving construction method of data compression model of satellite-borne wireless sensor network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150278441A1 (en) * | 2014-03-25 | 2015-10-01 | Nec Laboratories America, Inc. | High-order semi-Restricted Boltzmann Machines and Deep Models for accurate peptide-MHC binding prediction |
| CN107634937A (en) * | 2017-08-29 | 2018-01-26 | 中国地质大学（武汉） | A kind of wireless sense network data compression method, equipment and its storage device |
| CN107634943A (en) * | 2017-09-08 | 2018-01-26 | 中国地质大学（武汉） | A kind of weights brief wireless sense network data compression method, equipment and storage device |
-
2018
- 2018-12-26 CN CN201811603750.2A patent/CN109525598A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150278441A1 (en) * | 2014-03-25 | 2015-10-01 | Nec Laboratories America, Inc. | High-order semi-Restricted Boltzmann Machines and Deep Models for accurate peptide-MHC binding prediction |
| CN107634937A (en) * | 2017-08-29 | 2018-01-26 | 中国地质大学（武汉） | A kind of wireless sense network data compression method, equipment and its storage device |
| CN107634943A (en) * | 2017-09-08 | 2018-01-26 | 中国地质大学（武汉） | A kind of weights brief wireless sense network data compression method, equipment and storage device |
Non-Patent Citations (1)
| Title |
|---|
| JIANLIN LIU，FENXIONG CHEN: "Data Compression Based on Stacked RBM-AE Model for Wireless Sensor Networks", 《SENSORS 2018》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110446173A (en) * | 2019-07-31 | 2019-11-12 | 中国地质大学（武汉） | A kind of energy-efficient satellite-carried wireless Sensor Network data compression method |
| CN112153659A (en) * | 2020-08-21 | 2020-12-29 | 中国地质大学（武汉） | Efficient and energy-saving construction method of data compression model of satellite-borne wireless sensor network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109492822B (en) | Air pollutant concentration time-space domain correlation prediction method | |
| WO2021203242A1 (en) | Deep learning-based mimo multi-antenna signal transmission and detection technologies | |
| CN103900610B (en) | MEMS gyro random error Forecasting Methodology based on Lycoperdon polymorphum Vitt wavelet neural network | |
| Liu et al. | Automatic well test interpretation based on convolutional neural network for infinite reservoir | |
| Nie et al. | Network traffic prediction based on deep belief network and spatiotemporal compressive sensing in wireless mesh backbone networks | |
| CN101221213A (en) | Analogue circuit fault diagnosis neural network method based on particle swarm algorithm | |
| Chen et al. | Fog-based optimized kronecker-supported compression design for industrial IoT | |
| WO2021203243A1 (en) | Artificial intelligence-based mimo multi-antenna signal transmission and detection technique | |
| CN108734675A (en) | Image recovery method based on mixing sparse prior model | |
| Pei et al. | Data augmentation for rolling bearing fault diagnosis using an enhanced few-shot Wasserstein auto-encoder with meta-learning | |
| CN111210002B (en) | Multi-layer academic network community discovery method and system based on generation of confrontation network model | |
| CN104301728B (en) | Compression video acquisition and reconfiguration system based on structural sparse dictionary learning | |
| CN107634937A (en) | A kind of wireless sense network data compression method, equipment and its storage device | |
| Jiang et al. | Federated learning algorithm based on knowledge distillation | |
| CN109525598A (en) | A kind of fault-tolerant compression method of wireless sense network depth and system based on variation mixing | |
| CN107240136A (en) | A kind of Still Image Compression Methods based on deep learning model | |
| CN107634943A (en) | A kind of weights brief wireless sense network data compression method, equipment and storage device | |
| Du et al. | The Internet of Things as a deep neural network | |
| CN115620510A (en) | Traffic flow prediction method based on adaptive window attention extraction space-time dependence | |
| CN107910009A (en) | A kind of symbol based on Bayesian inference rewrites Information Hiding ＆ Detecting method and system | |
| CN102647354A (en) | End-to-end flow reconfiguration method in time-varying dynamic network | |
| CN116306780B (en) | Dynamic graph link generation method | |
| CN106021880A (en) | Jacket platform structure response computing method based on BP neural network | |
| CN117034189A (en) | Multi-source internet of things data fusion method | |
| Zhai et al. | Data reconstructing algorithm in unreliable links based on matrix completion for heterogeneous wireless sensor networks |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190326 |