基于特征合成的周期性备件需求预测方法Symbol`@@

Demand forecasting method for periodic spare parts based on feature synthesis

摘要:

针对工程机械备件需求预测准确性低的问题，提出一种新的基于特征合成的周期性维修备件需求预测方法. 定义等间隔备件需求样本集的相似度模型，采用优化算法确定最优备件需求周期长度，并利用回归模型建立各周期内的备件需求模型；提出基于特征合成的模型综合方法，借鉴物理力学中的矢量合成方法，将多个历史备件周期需求模型特征矢量合成新的特征矢量，利用新特征矢量还原获得最优的周期预测模型，该模型综合考虑了各个历史备件周期预测模型，使获得的备件周期预测模型具有更好的鲁棒性和泛化性. 采用人工数据和矿用圆环链的实际需求数据对该预测模型进行验证，实验结果表明，该模型具有良好的稳定性和准确性.

Abstract:

A novel approach for forecasting the periodic demand of spare parts based on feature synthesis is proposed to solve the problem of inaccurate prediction due to many influencing factors of the spare parts demand for construction machines and the demand cycle hard to be selected. The optimal demand cycle length is obtained with the optimization algorithm by defining a similarity measuring model of the spare parts demand sample sets under equal space, and the demand model for spare parts in every cycle period is built by the regression model. Then, a method is presented to integrate multiple cycle demand models of the spare parts into one according to the vector synthesis method in physics, so the optimal demand forecasting model for the spare parts with periodic pattern is obtained by reduction technology. The model synthetically considers the demand forecasting model for the spare parts in every historical cycle and it is great robust and generalized. The prediction model is verified by simulated datasets and the practical data of the demand of round link chains for mining, and experiment results prove that the model has good stability and accuracy.