Avaliação de técnicas de previsão de demanda para estimar produção de água de coco: o caso de uma agroindústria

Abstract

With the need to improve their performance and remain competitive in the market, companies are looking for tools to support their decision-making in the strategic sphere. Thus, the demand forecast has become a great ally and has been very useful for those who understand the importance of living together and being prepared for the fluctuations in the company's demand needs over time. This work brought a quantitative and computational supply & demand approach in order to analyze which method is best suited to the reality of a coconut water bottling company, located in Petrolina, in the São Francisco Valley region of Pernambuco. The data used were on the amount of fruit (green coconut) purchased and which is affected by seasonality. Four models were selected based on error parameters: Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Square (RMSE) and Scalar Mean Absolute (MASE). Among the forecast models used, five performed better based on the error parameters. The selected models were: Short-Term Load Forecasting (STLF) (MAPE = 39,44034; MAE = 100677.00; RMSE = 124480.40; MASE = 0.4908916); Neural autoregression (MAPE = 32.91682; MAE = 106726.00; RMSE = 130103.90; MASE = 0.5203861); Multiplicative Holt-Winters (MAPE = 41.90111; MAE = 101470.20, RMSE = 134650.50; MASE = 0.4947594); Holt-Winters Additive (MAPE = 44,7642, MAE = 107720.00; RMSE = 136886.30; MASE = 0.5252326); Arima by Box & Jekins (MAPE = 38.91704; MAE = 108053.50; RMSE = 142522.40; MASE = 0.5268587). It is worth nothing that production was affected in 2020 by the COVID-19 pandemic. Nevertheless, the study showed a good result and the prediction model that had the best performance for the Neural Autoregression with an error of 11.86% within over 5 moths accumulated of calculation, which represents a difference of 202,604 coconut units between the predicted model and the actual data. Sumary, the company in question will have more assertiveness in decision making in relation to its production, which will result in a better financial performance.