RU2014116187A

RU2014116187A - METHOD FOR IDENTIFICATION OF GASOLINE COMPONENTS AND DETERMINATION OF ITS COMPOSITION IN REAL TIME

Info

Publication number: RU2014116187A
Application number: RU2014116187/08A
Authority: RU
Inventors: Владимир Александрович Соловьев; Анна Алексеевна Щербакова
Priority date: 2014-04-22
Filing date: 2014-04-22
Publication date: 2015-10-27
Also published as: RU2568330C2

Abstract

Способ идентификации компонентов бензина и определения его состава в режиме реального времени, включающий создание базы данных спектральных коэффициентов поглощения тестовых компонентов, измерение спектральных коэффициентов поглощения тестовых компонентов и бензина в технологических линиях, отличающийся тем, что сигналы о спектральных коэффициентах поглощения тестовых компонентов и бензина суммируются нейронами идентификации, число которых равно или больше числа компонентов в бензине, и нейроном состава, с соответствующими синаптическими коэффициентами, величины и знаки которых, а также значения активации нейронов устанавливаются по образцам тестовых компонентов и бензинов с использованием алгоритма обратного распространения ошибки, в котором минимизируются целевые функции идентификации Uи состава В:причем идентификацию j-компонентов производят по минимумам целевых функций Uи активному состоянию нейронов искусственной нейронной сети идентификации, изменяя при этом коэффициенты синаптических связей ω, а значения активации нейронов определяют в соответствии с функцией активацииT=e,при этом концентрацию компонентов в бензине определяют по минимуму целевой функции состава и активному состоянию нейрона искусственной нейронной сети состава, изменяя коэффициенты синаптических связей ω, а значения активации нейрона состава вычисляют в соответствии с функцией активацииR=е,и при активном состоянии нейрона состава принимают значения концентрации компонентов в смеси cчисленно равным значениям коэффициентов синаптических связей ω,где k(λ) - спектральные коэффициенты поглощения тестовых компонен�A method for identifying gasoline components and determining its composition in real time, including creating a database of spectral absorption coefficients of test components, measuring spectral absorption coefficients of test components and gasoline in production lines, characterized in that signals about spectral absorption coefficients of test components and gasoline are summed by neurons identification, the number of which is equal to or greater than the number of components in gasoline, and a neuron composition, with the corresponding synap coefficients, the magnitudes and signs of which, as well as the values of activation of neurons, are determined from samples of test components and gasolines using the error back propagation algorithm, in which the target identification functions U and composition B are minimized: moreover, the identification of j-components is performed by the minima of the objective functions U and the active state neurons of an artificial neural network identification, while changing the coefficients of synaptic connections ω, and the activation values of neurons are determined in accordance and with the activation function T = e, while the concentration of components in gasoline is determined by the minimum of the objective function of the composition and the active state of the neuron of the artificial neural network of the composition, changing the coefficients of synaptic connections ω, and the activation values of the neuron of the composition are calculated in accordance with the activation function R = e, and at the active state of the composition neuron takes the values of the concentration of components in the mixture numerically equal to the values of the coefficients of synaptic connections ω, where k (λ) are the spectral absorption coefficients of the test components

Claims

A method for identifying gasoline components and determining its composition in real time, including creating a database of spectral absorption coefficients of test components, measuring spectral absorption coefficients of test components and gasoline in production lines, characterized in that signals about spectral absorption coefficients of test components and gasoline are summed by neurons identification, the number of which is equal to or greater than the number of components in gasoline, and a neuron composition, with the corresponding synap coefficients, the magnitudes and signs of which, as well as the activation values of neurons, are established from samples of test components and gasolines using the error back propagation algorithm, in which the objective identification functions U _j and composition B are minimized:

moreover, the identification of j-components is carried out according to the minima of the objective functions U _j and the active state of the neurons of the artificial neural identification network, while changing the coefficients of synaptic connections ω _ij , and the activation values of the neurons are determined in accordance with the activation function

T _j = e ^-Ujβε ,

the concentration of components in gasoline is determined by the minimum of the objective function of the composition and the active state of the neuron of the artificial neural network of the composition, changing the coefficients of synaptic connections ω _ij , and the activation values of the neuron of the composition are calculated in accordance with the activation function

R = e ^-Bαε ,

and with the active state of the composition neuron, the values of the concentration of components in the mixture c _{j are} numerically equal to the values of the coefficients of synaptic connections ω _ij ,

where k _jet (λ _i ) are the spectral absorption coefficients of the test components in the database;

k _j (λ _i ) - the measured spectral absorption coefficients of the components in the production lines;

k _cm (λ _i ) - the measured spectral absorption coefficients of gasoline in the production lines;

λ _i - wavelengths at which the spectral absorption coefficients are measured;

T _j - activation functions of the artificial neural network identification;

j is the number of components in gasoline;

R is the activation function of the artificial neural network composition;

α, β are coefficients that take into account the slope of the activation functions;

ε is the measurement error of the spectral absorption coefficients.