Scholarlink

<p>This study developed price forecasting models for the CEPEA/ESALQ White Crystal Sugar and Hydrous Ethanol Fuel indicators, incorporating variables related to sugarcane supply and climatic conditions. Machine learning models such as Long Short-Term Memory (LSTM), Transformer, and Multilayer Perceptron (MLP) were used, along with the statistical Autoregressive Integrated Moving Average (ARIMA) model, all incorporating exogenous variables. Among these variables are estimates from the Modular Agronomic Simulator for Sugarcane (SAMUCA), data from the National Supply Company (Conab), actual production figures, and a climate indicator constructed from temperature, precipitation, and solar radiation data provided by NASA POWER. Results indicate that models using SAMUCA production estimate exhibited accuracy comparable to those based on historical production data (used as a benchmark), with minor variations in error metrics. The MLP model achieved the best performance for sugar (MAPE of 3.84%), while LSTM was most effective for ethanol (MAPE of 1.87%). Machine learning techniques outperformed traditional methods in capturing seasonal, climatic, and nonlinear patterns. The proposed approach enables price forecasting in advance of the harvest and allows for monthly updates, offering a strategic tool for market stakeholders. The model is also adaptable to other crops and regions with limited production data availability.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Using deep learning models and exogenous production variables to forecast prices in the Brazilian sugarcane sector

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月6日

查看全文
  • Fernanda Cigainski Lisbinski,
  • Felipe André Oliveira Freitas,
  • Fabio Ricardo Marin,
  • Heloisa Lee Burnquist

摘要

This study developed price forecasting models for the CEPEA/ESALQ White Crystal Sugar and Hydrous Ethanol Fuel indicators, incorporating variables related to sugarcane supply and climatic conditions. Machine learning models such as Long Short-Term Memory (LSTM), Transformer, and Multilayer Perceptron (MLP) were used, along with the statistical Autoregressive Integrated Moving Average (ARIMA) model, all incorporating exogenous variables. Among these variables are estimates from the Modular Agronomic Simulator for Sugarcane (SAMUCA), data from the National Supply Company (Conab), actual production figures, and a climate indicator constructed from temperature, precipitation, and solar radiation data provided by NASA POWER. Results indicate that models using SAMUCA production estimate exhibited accuracy comparable to those based on historical production data (used as a benchmark), with minor variations in error metrics. The MLP model achieved the best performance for sugar (MAPE of 3.84%), while LSTM was most effective for ethanol (MAPE of 1.87%). Machine learning techniques outperformed traditional methods in capturing seasonal, climatic, and nonlinear patterns. The proposed approach enables price forecasting in advance of the harvest and allows for monthly updates, offering a strategic tool for market stakeholders. The model is also adaptable to other crops and regions with limited production data availability.