基于不同时间尺度的LSTM模型下产品订单需求量的预测

Prediction of Product Order Demand Based on LSTM Models with Different Time Scales

本文使用LSTM模型对日、周、月三个时间尺度下的产品需求量进行预测。结果表明,采用日尺度模型时,由于时间序列过长,模型无法达到收敛状态,在构建模型时应缩短时间序列;而使用周尺度模型时,损失函数逐渐收敛于横轴,训练效果较为显著;月尺度模型仍有提升空间,需要进一步训练以完全收敛。不同时间尺度会对预测精度产生影响,很大或很小的时间尺度可能导致预测误差增加。

This article uses LSTM models to forecast the demand for products in four sales regions at daily,weekly,and monthly time scales.The results indicate that when using the daily scale model,the time interval is too long for the model to converge,suggesting the need to shorten the interval when building the model.On the other hand,when using the weekly scale model,the loss function gradually converges to the x-axis,indicating a more significant training effect.The monthly scale model still has room for improvement and requires further training for complet e convergence.The choice of different time scales has an impact on the accuracy of the predictions,as very short or very long time scales can increase prediction errors.