A Novel Spatial-temporal Clustering-based Forecasting Method for Blast Furnace Gas Holder Level in Steel Industry

Abstract

The scheduling of blast furnace gas (BFG) system has an important influence on the energy consumption level of a steel plant. Accurate prediction of the BFG holder level could effectively reflect the state of the system, which is of great significance for the scheduling work. In this paper, a novel spatial-temporal clustering-based forecasting method was proposed. In order to describe the spatial features, a conversion method for transforming the correlation to relative distance was proposed, which could find the maximum correlation time delay by an improved grey correlation and take the delay as the relative distance. Moreover, a deep learning forecasting model based on a gated recurrent unit network was designed to fully exploit the spatial-temporal features, and the spatial clustering was adopted to extract major features, which further improved the forecasting accuracy. Experiment results showed that, compared with the currently existing methods, the proposed method had higher forecasting accuracy and was effective for BFG gas holder level forecasting.