Prediction Model Of Total Consumption Of Crude Steel Based On BP Neural Network And Support Vector Machine

At present,in the background of supply-side reform,the steel industry is facing the depth of structural adjustment.But for enterprises,it's also a very good development opportunities,accurate steel consumption data will provide a scientific support to China's steel overcapacity adjustment,but also for China's steel industry structural adjustment.In this paper,the main research goal is to establish a forecast model on the indicator,the total consumption of crude steel.According to the total consumption of crude steel caused by many complex and interactional factors,first,gray correlation analysis is used to reduce the dimensions,find the gray correlation coefficient between each indicator and the total consumption of crude steel,and select the indicator according to the correlation coefficient.Second,a intelligent simulation system model based on BP neural network is established,which is the relationship between the total consumption of crude steel and the influence factors.It is found that the predicted value of the model can be consistent with the expected value,but the forecast error of for individual points is relatively large,this shows that the forecast model is need to improve.Then,the simulation model of total consumption forecast of crude steel based on support vector machine is established.This model is better than BP model.Because of the BP neural network using gradient descent method,thus it inevitably exists the problems or defects,such as networking training slow,sensitive to the initial weights and thresholds,easy to fail into local minimum point and so on.While genetic algorithm has the advantages of global optimization capability,the initial value independence,a faster convergence rate,and preventing it from trapping at local minima.So with the genetic algorithm to optimize BP neural network initial weights and thresholds,this article establish a prediction model based on GA-BP network for the total consumption of crude steel.The optimized GA-BP network is superior to the non-optimized BP neural network model in prediction accuracy.This shows that the GA-BP network is more suitable for the forecast of the total consumption of crude steel.However,due to the GA-BP algorithm is built on the basis of BP network,if BP network is not good,that is,selection of the training samples or determination of hidden nodes numbers is unreasonable,the GA is not well optimized.In addition,the BP neural network is based on a large sample,while samples of the total consumption of crude steel are limited.At the same time,in order to avoid artificially setting the parameters blindly,uses genetic algorithms similarly to optimize relevant parameters of support vector machine,and establish a prediction model based on GA-SVM for the total consumption of crude steel.The experimental results show that the GA-SVM prediction model is the best,the average relative error is only 1.61308%.In summary,the GA-SVM forecast model established in this paper is satisfactory for the total consumption of crude steel,and it is possible to predict the total consumption of crude steel in practical work.This study can provide a reference for the government's macro-control in the steel market,and also provide a basis for traders to deal in the capital market.