基于多头注意力机制和长短时记忆网络的布匹产量预测模型

打开文本图片集

中图分类号：TS195.644 文献标志码：A 文章编号：2097-2911-（2025）05-0073-09

Abstract： With the development ofChina's textile industry，the prediction of fabric production is facing increasingly complex challenges.Traditional prediction models cannot effectively capture the impact of different time periods on production forecasts，resulting in inaccurate prediction results and a long model convergence time. To address this problem，a fabric production prediction model （MHAM-LSTM） is proposed based on a combination of a multi-head attntion mechanism （MHAM） and a long short-term memory network （LSTM）. By introducing the MHAM mechanism，the model can assign different weights to fabric production data in diferent time periods， thereby effectively sreening out time points that have a greater impact on future production forecasts.The introduction of the MHAM mechanism helps to reduce the training time of the model and improve the accuracyof the prediction.The LSTM network further optimizes the model's performance in time series data and can capture long-term dependencies over a long period of time.Experimental results show that the model with the MHAM mechanism introduced has achieved a significant improvement in prediction accuracy compared to the traditional LSTM model，with an R2 coefficient of up to 0.995.In the monthly prediction of textile production，the SE-Bi-LSTM model has the smallest error compared to the traditional BPand RNN models， with an average error of 0.22% ，areduction of 60.71% and 53.19% respectively.

Keywords： multi-head atention mechanism;long-short term memory network; fabric; production forecast; evaluationindex

随着全球纺织行业智能化转型的加速，布匹产量预测作为生产计划制定的核心环节，能够为原材料采购、设备调度、库存管理及订单交付提供关键决策依据。（剩余10382字）