Missing Data Imputation Based On An Improved Generalized-Trend-Diffusion For Gas Flow Of Steel Industry

As a resource-intensive industry, the steel industry consumes large quantities of energy resource, in which the byproduct gas has been the focused energy media of energy scheduling. The scheduling needs the support of complete energy flow data, however the phenomenon of data missing, especially consecutive missing, often occurs due to the events such as data collector failures, transmission errors, information storage errors, etc. At present the workers of the domestic steel plant use simple manual based method to solve this problem. However, such imputation accuracies cannot meet the requirement of the gas balance scheduling, and even might result in a complete failure for the scheduling and an excessive amount of gas emission. Thus, the data missing of steel industry is related to not only the economic efficiency of enterprises but also the environment protection.The blast furnace gas (BFG) system of Shanghai Baosteel is taken as the research background of this study. The original Generalized-Trend-Diffusion (GTD) is improved to enlarge the implicated knowledge of the sampled data via replacing the triangle membership function with the Gaussian one. And the enlarged data sample is treated as the input of a neural network with back propagated algorithm (BP). In order to make the parameters of this BP networks and the imputation sequence more suitable to the practice, a series of comparative experiments are carried out.. These experiments indicate that the improved GTD is superior to some other popular methods when dealing with the gas flow imputation of steel industry, especially the sort of problems with data properties of consecutively missing and small samples.