Abstract: Due to the different production conditions and impact parameters, looms have great differences in the actual weaving efficiency and loom warp-out times from each other. Aiming to solve the problem that the deviation between the theoretical calculation value and the actual value of loom warp-out time is too large when calculating it by using the pre-set static parameters of planned production, a loom warp-out time prediction method based on LSTM (Long Short Term Memory)  recurrent neural network is proposed. Based on the analysis of the factors affecting the loom warp-out time from three aspects: the warp and weft stop of the loom, the working efficiency of personnel and the variety of cloth processed, a data set of loom production with time series characteristics is constructed. The prediction of the model in the whole life cycle of the weaving axis is dynamically adjusted by setting the time schedule coefficient, and the performance of the model is optimized from the two aspects of loss degree and training time. Finally, eight groups of experimental data are used to verify the reliability of the model. The experimental results show that when it is 30 hours to 6 hours before the warp-out time of machine prediction, the average error range between the predicted value of the model and the actual value is 0.84 h to 1.52 h, which meets the requirements of actual production.

Key words:  , loom warp-out time, LSTM recurrent neural network, time series, warp and weft stop, weaving axis

摘要： 不同织机由于生产情况和影响参数各异，实际的织布效率和了机时间也存在着很大的差别。针对利用预先设定好的计划生产静态参数对织机了机时间进行计算时，存在理论计算值与实际织机了机时间偏差过大的问题，提出了一种基于长短时记忆(Long short term memory，LSTM)循环神经网络的织机了机预测方法。从织机经纬向停车情况、人员工作效率、加工布匹品种3个方面出发，分析影响织机了机时间的各类因素，构建了具有时间序列特性的织机生产情况数据集。通过设置时间进度系数动态调整模型在织轴整个生命周期内的预测情况，并从损失程度和训练耗时两方面考虑对模型性能进行优化。最后，利用8组实验数据对模型的可靠性进行验证。结果表明：模型在了机预测截止时间的前30 h至前6 h，模型的预测结果值与实际值之间的平均误差范围为0.84 h至1.52 h，满足对实际生产时的所需指标要求。

关键词: 织机了机, LSTM循环神经网络, 时间序列, 经纬向停车, 织轴