Scholarlink

<p>Agricultural price time series often exhibit complex, nonlinear, and nonstationary patterns, posing significant challenges for conventional forecasting methods. Although Empirical Mode Decomposition (EMD) is widely used for analyzing such data due to its adaptive and data-driven nature, it is inherently limited in handling bivariate data, as it cannot effectively capture the interdependencies between paired signals. To overcome this limitation, this study introduces a novel forecasting model that combines Bivariate Empirical Mode Decomposition (BEMD) with long short-term memory (LSTM) networks for interval-valued agricultural price forecasting. The model utilizes daily minimum and maximum potato prices from the Agra market, obtained from the ‘Agmarknet’ portal (<a href="https://agmarknet.gov.in">https://agmarknet.gov.in</a>), to construct an interval-valued bivariate time series. This series is transformed into a complex-valued signal, where the minimum (lower bound) and maximum (upper bound) prices represent the real and imaginary components, respectively. BEMD is then applied to decompose the signal into a set of Intrinsic Mode Functions (IMFs) and a residual component, each capturing distinct frequency characteristics. The real and imaginary parts of these decomposed components are extracted and modelled independently using LSTM networks. The individual forecasts of the IMFs and the residue are subsequently combined to produce the final interval forecasts. Comparative analysis reveals that the proposed BEMD–LSTM model significantly outperforms traditional EMD-based methods, as measured by Theil’s U statistic, Interval Mean Squared Error (IMSE), and Interval Mean Absolute Error (IMAE). These results underscore the enhanced capability of BEMD-based frameworks in capturing the dynamics of interval-valued agricultural price data, offering superior forecasting accuracy and robustness.</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

BEMD–LSTM model for interval forecasting of agricultural prices

  • 期刊论文

  • 发表日期：

    2026年3月12日

查看全文
  • Rounak Kumar,
  • Girish Kumar Jha,
  • Rajeev Ranjan Kumar,
  • Kapil Choudhary,
  • A. Praveenkumar,
  • Rajender Parsad,
  • S. B. Lal,
  • Samarth Godara

摘要

Agricultural price time series often exhibit complex, nonlinear, and nonstationary patterns, posing significant challenges for conventional forecasting methods. Although Empirical Mode Decomposition (EMD) is widely used for analyzing such data due to its adaptive and data-driven nature, it is inherently limited in handling bivariate data, as it cannot effectively capture the interdependencies between paired signals. To overcome this limitation, this study introduces a novel forecasting model that combines Bivariate Empirical Mode Decomposition (BEMD) with long short-term memory (LSTM) networks for interval-valued agricultural price forecasting. The model utilizes daily minimum and maximum potato prices from the Agra market, obtained from the ‘Agmarknet’ portal (https://agmarknet.gov.in), to construct an interval-valued bivariate time series. This series is transformed into a complex-valued signal, where the minimum (lower bound) and maximum (upper bound) prices represent the real and imaginary components, respectively. BEMD is then applied to decompose the signal into a set of Intrinsic Mode Functions (IMFs) and a residual component, each capturing distinct frequency characteristics. The real and imaginary parts of these decomposed components are extracted and modelled independently using LSTM networks. The individual forecasts of the IMFs and the residue are subsequently combined to produce the final interval forecasts. Comparative analysis reveals that the proposed BEMD–LSTM model significantly outperforms traditional EMD-based methods, as measured by Theil’s U statistic, Interval Mean Squared Error (IMSE), and Interval Mean Absolute Error (IMAE). These results underscore the enhanced capability of BEMD-based frameworks in capturing the dynamics of interval-valued agricultural price data, offering superior forecasting accuracy and robustness.