摘要
小样本条件下的销量预测,由于样本数量较少导致特征多样性降低,从而影响模型泛化性和准确性。以汽车产业链平台上“HB”企业的整车销售数据为研究对象,提出一种小样本下基于业务流程执行状态的目标特征细分处理方法,该方法可对实时目标值划分更细的粒度。同时,将特征加权引入到整车需求量预测问题上,并构建基于XGBoost+LightGBM+LSTM的组合预测模型。最终实验表明,相比于单一预测模型,最优加权组合模型的预测效果具有更高的准确率。
The sales forecast under the condition of small samples,due to the small number of samples,reduces the feature diversity,which affects the generalization and accuracy of the model.Taking the vehicle sales data of HB companies on the automobile industry chain platform as the research object,a small sample-based target feature segmentation processing method based on business process execution status is proposed.This method can divide real-time target values into finer granularity.At the same time,feature weighting is introduced to the problem of vehicle demand forecasting,and a combined forecasting model based on XGBoost+LightGBM+LSTM is constructed.The final experiment shows that,compared with a single prediction model,the prediction effect of the optimal weighted combination model has a higher accuracy rate.
作者
焦笑
任春华
司佳顺
Jiao Xiao;Ren Chunhua;Si Jiashun(Manufacturing Industry Chains Collaboration and Information Support Technology Key Laboratory of Sichuan Province,Southwest Jiaotong University,Chengdu 611756;Beijing Machinery Industry Automation Research Institute Co.,Ltd.,Beijing 100120)
出处
《现代计算机》
2021年第23期36-42,共7页
Modern Computer
基金
国家重点研发计划资助项目(2017YFB1401400)。
关键词
特征细分加权
小样本学习
LSTM模型
组合预测模型
汽车产业链
feature subdivision weighting
small sample learning
lstm model
combination prediction model
automobile industry chain