Research And Implementation Of Steel Demand Forecasting System Based On Convolution And Self-Attention

The iron and steel industry is an important industry in China’s economic development and a significant symbol for measuring China’s economic level and industrialization.With the rapid development of the Industrial Internet,China is also actively participating in the globalization wave of new manufacturing changes.The purchasing decisions of the traditional steel industry often have the following problems:1.There are a large number of steel grades.If each steel grade is modeled separately,the system calculation cost and data storage cost will increase sharply,and if all steel grades are modeled uniformly It is difficult to obtain accurate prediction results;2.There are few available attributes.The characteristics of industrial big data are that the data is scattered and the information contained is complex.The performance of the steel demand data is that the known attributes are complex,but the useful attributes are scarce;3.Too much dependence Manual,resulting in unscientific decision-making process,low prediction accuracy and high labor cost.This paper proposes a steel grade demand forecasting system for the Industrial Internet,the core of which is a two-stage many-to-many time series forecasting algorithm based on convolution and self-attention.In the first stage,the apriori algorithm is used to extract high-RRSEelation steel species sets,and in the second stage,temporal convolutional networks with different window sizes are used to extract the global and local temporal features of the time series sets respectively.The pooling layer of the product network is replaced by a strided convolutional layer,and then the self-attention stacking module is used to learn the interrelationship of the time series of different steel grades.The forecast results are combined to obtain the final demand forecast value.The users of the system are the staff of the purchasing department of the steel industry,which has low requirements for the user’s professional knowledge and purchasing experience,and can predict the demand of multiple steel grades at the same time,allowing the staff to manage the data of different steel grades and make purchasing decisions.Realize networking and intelligence.The steel grade demand forecasting system designed and implemented in this paper relies on the industrial Internet intelligent cloud collaboration platform of the national key scientific research project,and aims to provide an intelligent management and decision-making model for the steel industry procurement department.After many experiments,the algorithm proposed in this paper can meet the prediction requirements.The test results of the whole system show that each functional module can achieve the expected effect and can be used online.