Forecasting blast furnace gas production and demand through echo state neural network-based models: Pave the way to off-gas optimized management

Highlights: • Forecasting blast furnace gas volume flowrate and heating power by ESNs. • Prediction of blast furnace gas demand by hot blast stoves through ESNs. • Low prediction errors obtained during the test of the models with real data. • Inputs related to the scheduling of the process are fundamental for the models. • Low computational burden is required by the models during training stage. -- Abstract: The efficient use of resources is a relevant research topic for integrated steelworks. Process off-gases, such as the ones produced during blast furnace operation, are valid substitutes of natural gas, as they are sources of a considerable amount of energy. Currently they are recovered, for instance, by using in hot blast stoves but sometimes part of such gas is flared due to non-optimal management of such resource. In order to exploit the off-gases produced in an integrated steelworks, the interactions between gas producers and users in the whole gas network need to be considered. The paper describes two models exploited by a Decision Support Tool that is under development within a European project. Such models forecast, respectively, the blast furnace gas amount and its heating power by obtaining an error between 1.6 and 6.9% in a time horizon of 2 h and the blast furnace gas demand by hot blast stoves by giving a prediction error between 5.0 and 12.1% in the same time horizon. The forecasted values of blast furnace gas production and main demand allow a continuous optimal planning of the blast furnace gas usage according to its availability and to the needs in the steelworks, by avoiding losses of a valuable secondary resource and related emissions.