Processing of experimental data of coking process using neural network

Abstract

Oil conversion ratio is a major index of refinery efficiency. Thermal refining processes are the most effective and economically attractive processes as compared with catalytic processes. Delayed coking process increases the oil conversion ratio sufficiently. It is waste-free thermal process. In this paper comparative analysis of processing of statistical data methods is given. Two methods of statistical data processing are given: regression analysis and data processing by neural networks. Regression analysis is the classical method of data processing based on the compiling of the regression equation and selection of the regression coefficients. Data processing by neural networks based on algorithms that take place in the human brain between biological neurons. Neuron in an artificial neural network is a mathematical processing center with input and output data, synaptic weights, bias, adder and the activation function. The main data processing stages of each method are shown. Using neural network, processing of experimental data of coking process is done. Heavy catalytic cracking gasoil, heavy pyrolysis resin and vacuum residue are used as a feed for coking process to produce coke with low sulfur content. A different ratio of these components in the raw material mixture and the pressure in the reactor are variable parameters of the process. Obtained neural network adequately describes the coking process. Comparing the results of the two methods neural network give more a dense distribution of the calculated data with the experimental values.