基于ARIMA-LSTM的产品需求预测方法

Product Demand Prediction Method Based on ARIMA-LSTM

摘要: 企业产品的需求预测并非完全的线性或非线性问题，单一的预测模型的预测结果精度低，可靠性差。针对该问题，综合ARIMA和LSTM模型的优势，提出一种ARIMA-LSTM混合预测模型。为获得更高的模型预测精度，利用不同的方法对ARIMA和LSTM模型进行权值分配，选择预测精度更高的混合模型。利用某生产厂的需求数据进行实例分析，结果表明不同权值分配方法下的混合模型精度不同，由相对误差倒数法确定权值的混合模型具有更高的预测精度，该结果为后续的企业需求预测以及制定生产计划提供科学依据。

Abstract: The demand prediction of enterprise products is not a complete linear or nonlinear problem, and the prediction results of a single prediction model have low accuracy and poor reliability. To ad-dress this issue, a hybrid ARIMA LSTM prediction model is proposed by combining the advantages of ARIMA and LSTM models. To achieve higher model prediction accuracy, different methods are used to assign weights to ARIMA and LSTM models, and a hybrid model with higher prediction ac-curacy is selected. Using demand data from a certain production plant for example analysis, the re-sults show that the accuracy of mixed models varies under different weight allocation methods. The mixed model determined by the reciprocal method of relative error has higher prediction accuracy. This result provides a scientific basis for subsequent enterprise demand prediction and production planning.