INTEGRATED APPROACH FOR THE OPTIMIZATION OF THE CUTTING STOCK PROBLEM WITH USABLE LEFTOVERS AND DEMAND FORECASTING VIA MACHINE LEARNING

Industrial production faces significant challenges in planning its processes, where inadequate decisions, such as inefficient inventory management or improper use of materials, can lead to substantial losses. These inefficiencies compromise a company’s profitability and the environment due to excessive resource waste and unnecessary consumption of raw materials. In this context, this work proposes an integrated approach aimed at optimizing industrial production by combining two core elements: the development of predictive Machine Learning models based on recurrent neural networks of the LSTM (Long Short-Term Memory) type for demand forecasting, and the solution of the Cutting Stock Problem with Usable Leftovers (CSPUL), which minimizes production losses through the strategic reuse of leftover materials (scraps).

The LSTM model, a recurrent neural network (RNN) designed for sequential data, is well-suited for demand forecasting because it can learn long-term dependencies in time series. In this study, the LSTM model was trained on an artificial dataset that simulates the problem scenario, containing variables such as Client, Warehouse, Date, Price, and Demand. The variable Client represents the end consumer, with consumption profiles defined by preferences for item categories (Medium or Large). The Warehouse corresponds to the production site, responsible for controlling the stock of leftovers and managing the production capacity needed to meet demand. The variable Date captures temporal variation in the data, enabling dynamic analysis over time. The Price, which fluctuates periodically, directly impacts on Demand: price increases reduce consumption, while price decreases increase it, reflecting customer sensitivity to the product cost.

The demands predicted by the LSTM model are integrated into the solution obtained from a model in the literature for the CSPUL. This optimization problem aims to determine the best combination of cutting patterns for stock materials, to produce items in various sizes and quantities while minimizing waste. The distinguishing feature of the CSPUL lies in its ability to generate and reuse usable leftovers, increasing the process’s overall efficiency. The implementation of the model was carried out using the Gurobi Optimizer solver. Both standardized stock materials acquired by the company and previously stored leftovers are considered in the formulation, enabling the forecasted demand to be met more sustainably and efficiently.