Production optimization indexed to market demand using deep neural networks

Abstract Raw materials play an important role in the production of goods since their acquisition value has an important impact on final production. Controlling the variation of raw material costs is a huge challenge for companies. Predicting price variations in advance can lead to adjustments in strategic plans, benefiting the company. One of these strategic decisions is better planning of production capacity according to market demand, and adjusting equipment maintenance plans to ensure uninterrupted production during critical times while maintaining high quality standards. The present paper proposes a method to forecast paper pulp production indexed to stock market values using deep neural networks. The forecast leverages pulp production variables from paper presses combined with stock exchange variables, specifically ALTRI SGPS (volume and price), PSIALL-SHARE index, Consumer Price Index (CPI) and PCU32213221 (industry operating costs), as model inputs. The results show that stock market data significantly improves the forecasting model: the mean absolute percentage error (MAPE) was reduced from 26.55% to 5.08% (an 80% reduction from the baseline model without external variables), and R $$^{2}$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mmultiscripts> <mml:mrow/> <mml:mrow/> <mml:mn>2</mml:mn> </mml:mmultiscripts> </mml:math> improved from 0.42 to 0.86, with additional error reductions achieved through increased sampling rates. Graphical abstract