METHOD FOR NEURAL NETWORK CLUSTERISATION OF WIRELESS SENSOR NETWORK Russian patent published in 2015 - IPC H04W40/24 H04W84/18 G06N3/08 

FIELD: radio engineering, communication.

SUBSTANCE: method includes hierarchical division of notes into head cluster nodes (HCN) and slave HCN, and using node radio visibility data. The structure of the entire wireless sensor network (WSN) is described by a graph of the energy visibility of WSN nodes, based on which a visibility energy matrix is constructed, the matrix being multiplied by a reduction factor given as a percentage, and then transformed into an incidence matrix. Clusterisation is carried out using a Kohonen neural network, which is trained based on a constructive training method, where the input training data are a previously obtained incidence matrix; the number of neurons of the Kohonen network is set automatically based on the difference and similarity of input data on WSN nodes; the radius of sensitivity of Kohonen layer neurons is set in the range of 0.22 to 0.36. The visibility energy matrix of the WSN nodes is used for routing and enables to perform inter-cluster communication between HCN and intra-cluster communication within the slave of each HCN node.

EFFECT: automatic construction and maintaining operability of a network structure.

6 dwg