塔里木大学学报

红枣干果期货交易对稳定国内、尤其是新疆南疆的红枣干果价格发挥了重要的积极作用。开展红枣期货价格预测方法研究，有助于红枣产业利益相关者更好地组织生产、加工和投资管理。本研究采用3种价格预测模型对比分析了不同方法在红枣期货价格预测方面的综合性能差异，包括长短期记忆(LSTM)神经网络模型、支持向量回归(SVR)模型和误差反向传播(BP)神经网络模型。试验结果表明，LSTM价格预测模型在预测精度方面与SVR模型相比，均方根误差(RMSE)值降低了17.4%、平均绝对百分误差(MAPE)值降低了25%;与BP模型相比，RMSE值降低了12.8%、MAPE值降低了33.3%;在年度价格预测方面，LSTM模型取得了更佳的性能表现，尤其在提前5日预测红枣期货价格变动趋势时优势明显。基于LSTM的红枣干果价格预测模型可对红枣期货价格预测发挥辅助决策作用。

The trading of dried jujube futures has significantly contributed to stabilizing prices, particularly in Southern Xinjiang. Examining methods for predicting jujube futures prices can aid stakeholders in the industry to streamline production, processing, and investment strategies. This paper assesses three prediction models for jujube futures prices: long short-term memory(LSTM) neural network, support vector regression(SVR), and back propagation(BP) neural network. Results indicate that compared to SVR, the LSTM model reduces the root mean square error(RMSE) by 17.4% and the mean absolute percentage error(MAPE) by 25%. Compared to BP, LSTM reduces RMSE by 12.8% and MAPE by 33.3%. LSTM performs better for annual price prediction, especially in forecasting jujube futures trends five days ahead. The LSTM-based model can enhance decision-making in jujube futures price forecasting.