Method and device for reconstructing lost data of marine wireless sensor networks

Patent metadata

Field	Value
Publication number	US-12556906-B2
Application number	US-202318456478-A
Country	US
Kind code	B2
Filing date	Aug 26, 2023
Priority date	Jul 20, 2023
Publication date	Feb 17, 2026
Grant date	Feb 17, 2026

Abstract

Provided are a method and a device for reconstructing lost data of marine wireless sensor networks (MWSNs). The data reconstruction method includes following steps: establishing an initial topological structure of the MWSNs; using an improved hierarchical energy balance multipath (IHEBM) routing protocol to cluster network nodes; using an improved radial basis function neural network (RBFNN) to predict lost data of nodes in the cluster based on a clustering of the nodes; and using a centralized principal component analysis (PCA) method to compress and reconstruct data of a cluster head node in a process of data transmission from a cluster head to a ship base station.

First claim

What is claimed is:

1 . A method for reconstructing lost data of marine wireless sensor networks (MWSNs), comprising the following steps:
establishing an initial topological structure of the MWSNs; using a hierarchical energy balance multipath (HEBM) routing protocol to cluster network nodes, wherein goals of clustering MWSNs nodes by the HEBM routing protocol comprise: prolonging a service life of the MWSNs through a distributed energy consumption, realizing clustering in a changing iteration number and obtaining evenly distributed cluster heads; and specific steps of node clustering comprise a cluster head selection and a cluster member formation, wherein specific steps of the cluster head selection are as follows: calculating a probability of each node being selected as a cluster head candidate node in the MWSNs and broadcasting a first message to other nodes, wherein the first message comprises a node number and the probability of the node being selected as the cluster head candidate node; and indexes used for calculating the probability of a node i selected as the cluster head candidate node comprise: a distance between the node i and a ship base station; residual energies of the node i and neighbor nodes; mutual distances between the node i and the neighbor nodes; and a density of the neighbor nodes of the node i; if a probability of the node i being selected as the cluster head candidate node is greater than probabilities of all neighbor nodes j of the node i being selected as the cluster head candidate node, finally selecting the node i as the cluster head candidate node; if the probability of the node i being selected as the cluster head candidate node is less than probabilities of all neighbor nodes j of the node i being selected as the cluster head candidate node, sending a join message by the node i to all neighboring nodes having a greater probability of being selected as the cluster head candidate node and the node i being a member node of the cluster; and broadcasting a second message by each temporary cluster head selected as the cluster head candidate node, wherein the second message comprises a node number and a probability of the node being selected as the cluster head candidate node; when a temporary cluster head ch i receives the second message broadcast by a temporary cluster head ch i , the temporary cluster head ch i calculates a distance between the temporary cluster head ch i and the temporary cluster head ch i ; if the distance is greater than or equal to a threshold value d th , the second message is ignored; if the distance is less than the threshold value d th , and the probability of the node/being selected as the cluster head candidate node is less than the probability of the node i being selected as the cluster head candidate node, the temporary cluster head ch j becomes an ordinary node and sends a join message to the finally selected cluster head ch i ; using a radial basis function neural network (RBFNN) to predict lost data of nodes in the cluster based on a clustering of the nodes, wherein specific steps of predicting node lost data by the RBFNN comprise: obtaining an optimal structure of the RBFNN by training marine data in a data loss cluster, and accurately predicting node lost data by using the RBFNN after training; wherein when a training sample X k is input, a result y kj of a j-th output neuron of the RBFNN is:

y
kj

=

w

0
⁢
j

+

∑

i
=
1

I

w
ij

⁢
ϕ
⁢

(

X
k

,

X
i

)

,

j
=
1

,
2
,
…
,
J

wherein X i is a center of a basis function, k is a number of hidden layer nodes, w ij is a weight between an i-th hidden neuron and the j-th output neuron, and ϕ(X k , X i ) is a radial basis function; the optimal number k of the hidden layer nodes and the center X i of the basis function of the hidden layer of the RBFNN are determined by a clustering technology in the HEBM routing protocol; and
using a centralized principal component analysis (PCA) method to compress and reconstruct data of a cluster head node in a process of data transmission from cluster heads to the ship base station, wherein specific steps of compressing and reconstructing the cluster head node data by using the centralized PCA method are as follows:
obtaining initial marine perception data in a sampling period, wherein the initial marine perception data obtained by the cluster head comprises: a real marine sensing measurement value sent by a cluster member node to the cluster head, a predicted value generated at the cluster head and a real measurement value of the cluster head node; X=[x 1 , x 2 , . . . , x M ] T ∈ MxP represents a collected perceptual data set, M and P represent respectively a discretization time threshold and a number of nodes in the cluster, row vectors of the perceptual data set X corresponds to sensor nodes, and column vectors of the perceptual data set X corresponds to the measured values sent to the cluster head node by all cluster member nodes in a sampling period M;
centralizing the initial marine perception data to obtain a centralization matrix X mean ;
calculating a principal component, decomposing a covariance matrix

C
=

1
M

⁢

X
mean
T

⁢

X
mean

 into eigenvalues, sorting the calculated eigenvalues according to a size λ 1 ≤λ 2 ≤ . . . ≤λ p , and obtaining a component importance score matrix B:

B
=

X
mean

⁢
A

=

[

b
1

,

b
2

Assignees

• Univ Shanghai Maritime

Inventors

• Xian Jiangfeng
• Ma Junling
• Wu Huafeng
• Yang Yongsheng
• Zhang Yuanyuan
• Mei Xiaojun
• Chen Xinqiang
• Li Chaofeng
• Zhang Qiannan
• Liang Linian

Classifications

• H04W84/18

  Self-organising networks, e.g. ad-hoc networks or sensor networks · CPC title

• H04W16/26 Primary

  Cell enhancers {or enhancement}, e.g. for tunnels, building shadow · CPC title

• H04L45/24

  Multipath · CPC title

• G06N5/022

  Knowledge engineering; Knowledge acquisition · CPC title

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title