Scheduling Method And Its Application Of Blast Furnace Gas System Based On Depth Model

As a typical secondary energy system,blast furnace gas system is an important part of the production process of steel enterprises.The smooth operation of the blast furnace system is the guarantee of efficient production,energy saving and emission reduction.Therefore,modeling and scheduling blast furnace system through effective methods is of great significance to enterprise production.In this paper,the relevant problems of the blast furnace gas system are studied based on the deep learning methods,laying the foundation for forecasting and scheduling methods based on data analysis.To address the problem of missing data in the energy management system,a generative imputation method based on deep generative model was proposed.In this method,an interpretable input space for generative models was provided by fusing variational autoencoders and generative adversarial networks,and the generated sequence was matched with the target sequence by adding the reconstruction loss function to the target function.Meanwhile,to reduce the difficulty of data generation and improve the filling accuracy,this method filled in sub-sequences at multiple time scales by sequence decomposition.To address the problem of blast furnace gas system scheduling,within the framework of deep reinforcement learning,the problem was transformed into searching the best production state under a certain operating condition.Meanwhile,constraints are added to the reward function to transform the problem into unconstrained optimization problem.Moreover,in order to speed up convergence and improve algorithm stability,an experience based pre-training was added to the training session.To achieve scheduling evaluation,a multi-factor input short-term gas holder level prediction model based on gated recurrent neural network was proposed to evaluate and feedback the scheduling actions output by the strategy network in the scheduling model.In order to verify the operation effectiveness in the practical system,the proposed algorithms were simulated and verified by using actual field data.An industrial software developed on this basis validates the method proposed in this paper on the existing energy management system in the form of a network service process,and a control system with GUI was used to show the actual operation of the algorithm.It indicates that the algorithm in this paper can effectively deal with the problem of incomplete data in the blast furnace gas system,and can provide decision support for the scheduling of blast furnace gas.