machine learning and deep learning-based approach for time-series forecasting in steel industry: steel production forecasting

Steel making is a resource heavy industry, producing a lot of emissions and waste. researches show for every ton of steel, 0.55 ton of waste is produced. also, the logistics involved with extracting, processing, and transporting raw material into the plant and storing produced products and transporting them is costly and environmentally unfriendly. therefore, good production planning can lead to reduction of overproduction, waste, emissions, energy consumption and overall cost. therefore, providing the decision makers with forecasts of market indicators is very useful. production data is one such indicator which in itself embeds information on economic situations, demand, energy restraint among other things. one of main applications of artificial intelligence is time-series forecasting. in this research three machine learning and deep learning-based methods namely random forest, svr and gru are used for forecasting. also, industrial production index of us’s raw steel is used as dataset and proof of concept. it's shown that all models produce good results however gru has the lowest error and the best results with mae of 1.48 , mse of 3.97 and r2 of 0.870. lastly an index named pfi is introduced and formulated to show the fluctuation of steel production relative to historical averages.