A Scheduling Decision Support System For Blast Furnace Gas Based On Transfer Learning

The production process of steel industry consumes energy and pollutes seriously.How to use energy efficiently to reduce pollution emissions is an urgent problem for enterprises.Blast furnace gas(BFG)is an important secondary energy source in the production of steel enterprises.Optimizing the scheduling of gas system,effectively utilizing blast furnace gas and reducing gas emission are important means to realize energy saving and emission reduction in enterprises.In this paper,typical small-sample learning problems in blast furnace gas system are analyzed,and the method of gas tank modeling and adjustment point classification based on transfer learning is studied,so as to improve the optimal scheduling effect.Aiming at the problem of production condition change in the production process of blast furnace gas system,based on the similarity between different production conditions,a modeling method based on view ensemble transfer learning is proposed.Considering the problem of source domain transfer failure caused by variance inconsistency,a subspace extraction method based on domain mean-variance consistency is proposed.This method constructs multiple single-view auxiliary training set for multi-source domain and a multi-view training set is formed from target domain,which reduces the impact of negative transfer in single source domain on the overall output.Meanwhile,the accuracy of small sample modeling in target domain is improved by making full use of the multi-view algorithm.Considering the whole dynamic process of increasing sample size from zero for the new production condition,the domain ensemble function is constructed to weigh the domain error and the confidence of target domain model.In addition,according to the multi-output characteristics of blast furnace gas system,a multi-view multi-output least squares support vector machine regression algorithm(MVMO-LSSVM)is proposed to model multi-view data in target domain.As for the problem of identifying adjustment points of blast furnace gas system,a transfer learning method based on joint distribution adaptation is proposed on the basis of gas tank modeling and considering the similarity of gas systems between different enterprises.This method takes into account the similarity of edge distribution and conditional distribution between different domains data to estimate the data weight of auxiliary training set,avoiding the imprecise estimation of samples caused by label differences.And then the least squares support vector machine based on sample weighted classification algorithm is used to model,improved the generalization of a small amount of adjustment point data modeling.Based on the actual production data of a domestic iron and steel enterprise,the proposed method of tank level modeling and adjustment point classification are verified experimentally,and the results show that the proposed methods have a good effect on the scheduling of blast furnace gas system under the condition of small samples.Combined with the research method in this paper,the scheduling decision support system of gas is developed and applied to the actual production management of enterprises.The operation results show that the work of this paper has a good guiding role in the decision making of BFG dispatching,and is of great significance for enterprises to improve energy efficiency and realize green production.