Order Completion Time Prediction Based On Data Characteristics In The Equipment Manufacturing Industry

With the rapid development of science and technology,especially the advancement of computer and artificial intelligence technologies,discrete equipment manufacturing industries are undergoing a new round of industrial transformation,wherein intelligence,information,and industrialization technologies have been deeply integrated to accelerate the intelligent evolution.For discrete equipment manufacturing enterprises producing according to order requirements,accurately predicting order completion time is significant to ensure the timely delivery of orders to customers and company’s competitiveness.However,there exist many uncertain factors that impede predicting order completion time,since the complex environment of actual manufacturing processes would affect the final order completion time.We take the order completion time prediction problem in discrete equipment manufacturing industries as the research topic in this paper,and the main research contents are as follows:(1)To address the characteristics of small quantity,insufficient feature expression capability and contains noise in order data,an order completion time prediction algorithm based on data distribution and cascade forest is proposed.Order completion date prediction is considered as a classification problem based on the data characteristics,and a two-stage cascade forest model is designed.The original data is input to the first-stage cascade model,and the initial prediction result and the prediction probability distribution vector are output.By analyzing the probability distribution of order completion time,the prediction results and the prediction probability distribution vector are optimized and spliced to obtain a new feature vector with more expressive power.The new feature vector is spliced with the original data,and then the spliced data is input to the second-stage cascade model.Finally,the final prediction result is output by the second-stage cascade model.The experiments show that the average bias of the order completion time prediction of this algorithm is 1.11 days,and the accuracy is81.04%.(2)Aiming at the problem of unbalanced order data,a dynamic oversampling classification and prediction algorithm with a weighting mechanism is proposed.The algorithm uses the random forest as the base classification model and selects shapley values as feature weight,and then calculates the distance of samples based on the feature weight.Then,according to the distance and density of samples,using DBSCAN density clustering to classify the minority class samples into multiple types,and designing different sample synthesis strategies according to the distribution of different types of samples.Finally,a buffering mechanism is designed as the termination condition of dynamic oversampling,and the trained optimal classification model is output at the end of oversampling.The algorithm is experimentally demonstrated to exhibit better classification performance on several publicly available datasets.Compared with the original unused oversampling method on the imbalance order dataset,the proposed algorithm improves F-measure,G-mean,and AUC by 0.0423,0.0244,and 0.0196 respectively.This proves the effectiveness of the proposed algorithm.