Igor V. Denisov

Selection of optimal parameter of speed of training of neural network perceptron type

The problem of optimization of training of the neural networks perceptron type is considered. The method of the select of parameter of the speed of training based on combination of optimum parameters is offered: optimum parameter selected before the beginning training and the optimum parameter after the first cycle of training. The mathematical calculations and deductions of optimum parameter of the speed of training represented. The numerical modeling on an example of the neural networks perceptron type is realized. Is shown that the combination optimum parameter of the speed of training on a first step of training with optimum parameter of the speed of training received previously improves the result of training more than on 0,24% and the speed of training more than one and a half.

One-wire network standard for the fiber optic measuring network

The principle of optical information collection from the fiber-optical measuring network (FOMN) is submitted. The principle of functioning 1-Wire network standard lays in the basis of the implemented method. The represented device is a function block, intended for collecting optical information from 4 fiber-optical measuring line (FOML), converting optical information into digital signals and delivery of digital information about intensity of laser radiation into FOML. The device has a very small amount of elements, that considerably makes more cheap practical realization of such blocks, has small overall dimensions and good operating performances.

Synthesis of approximate algebraic and neural-like methods for the solution of the tomography problem

The approximate algebraic methods of the solution of a tomography problem of restitution of the performance of extended physical fields in matching with neural-like method of the solution of such problem are considered. The analyses of methods and results of modeling are made.

Organization of fiber optical temperature measuring system

The block diagram of the device intended for data processing organize from fiber-optical measuring network (FOMN), modeling and controlling parameters of the temperature field for FOMN is submitted. The principle of functioning 1-Wire netowrk standard lays in the basis of the device. The practical realization of this system allows to collect optical information from 15 fiber-optical measuring lines (FOML), formed the FOMN with packing 4x4, convert optical information into digital signals and delivery digital information about intensity of laser radiation into FOML. The part of modeling and controlling the parameters of the temperature field is necessary to form a matrix of connections of optical neural network.

Distributed physical field monitoring by using a photomatrix neural-like system

In the given paper principles of organization of neural-like system consisting of a matrix of photoelectric cells are represented. The practical realization of this system allows to obtain a parallel processing of an optical information for environmental physical field monitoring . A computer model of the feed-forward neural network with the hidden layer is developed to reconstruct physical field investigated by the fiber optic measuring system. The Gaussian distributions of some physical quantity are selected as learning patterns. Neural network is learned by error back-propagation using the conjugate gradient and coordinate descent minimization of deviation. Learned neural network reconstructs the two-dimensional scalar physical field with distribution having one or two Gaussian peaks.

Ternary influences on the fiber-optical measuring network

The purpose of the given work is further solution of actual fiber-optical tomography problem of spatial distribution reconstruction of the physical influences on the fiber-optical measuring networks. The problem of simultaneous reconstruction of the places and values of ternary influences on fiber-optical measuring network from 3×3 to n×m dimension is described. For discussion of this problem were used the algebraic methods for solution of the system of linear algebraic equations. As the tomography data the integrated data coming from the fiber-optical measuring lines, assembled according to the perpendicular stacking scheme on the fiber-optical measuring network were used.

Neural-like optoelectronic processing system

In the given article the method of optical information gathering from the fiber-optical measuring network with its subsequent processing is offered. In this method the algorithms of neural-like networks in computation process is introduced. Each sensitive area of the fiber-optical measuring line is associated with the own amplifier. Adjustment of amplifiers gain factors carries out modification of the weighting coefficients of the matrix of connections of the neural network. The training principles to external physical influences are represented. The selection of the type of the neural network for decision of the fiber-optical tomography problem of spatial distribution reconstruction has been considered.

Construction of the fiber-optical temperature measuring system

A fiber-optic amplitude sensor based on the effect of microbending-caused upsetting of total internal reflection intended for temperature monitoring is offered. The measuring breadboard is described for investigating the sensor characteristics. It is shown that several sensors of this type can be integrated in fiber-optic measuring lines to be used in distributed fiberoptic measuring networks. The characteristics of a fiber-optic measuring line composed of three fiber-optic microbending amplitude sensors are investigated. The fiber-optical measuring network on base fiber-optic microbending amplitude sensors with three-direction stacking of lines and dimension 4×4 is suggested. The fiber-optical measuring system for reconstruct the characteristics of distributed physical fields on developed fiberoptical measuring network is described. The results of reconstruction of two temperature influences with 46 and 74,5 Celsius degree value is represented.

Fiber-optical phase sensitive surface

The sensitive surface on the basis of fiber-optical measuring network with demodulation phase filters is offered. The purpose of the given work is further solution of actual fiber-optical tomography problem of spatial distribution reconstruction of the physical influences on the fiber-optical measuring networks. The problem of simultaneous reconstruction of the places and values of influences on fiber-optical measuring network from 4×4 dimension is described. For discussion of this problem were used the algebraic methods for solution of the system of linear algebraic equations with combinations of neural-like algorithms perceptron type. As the tomography data the integrated data coming from the fiber-optical measuring lines stacked on two and three directions on fiber-optical measuring network of researched area were used.

Adaptive information interchange system of the fiber-optic measuring networks with the computer

In the present paper the characteristics and opportunities of application of the system of parallel input-output of information from the fiber-optical measuring network into computer are considered. The system consists of two pars: on manframe and several expansion blocks. The first part is internal, is connected directly in the socket of the motherboard of the personal computer. It is designed for buffering system signals and development of cojmands of controlling by the system for input-output of signals into personal computer and signals generation onto expansion blocks. The second part is external, connects to the mainframe by means of cables. It designed for transformation of information from the fiber-optical measuring network into signalsof rthe mainframe and instrument settings adaptation. The analysis of speed of procesing of analog and digital data by system is presented. The possible schemes of use of the system for processing quasistationary and dynamic fields are considered.