Data-driven Multi-scale Modeling And Optimal Scheduling For Steam System

As one of the important secondary energy sources in steel enterprises,steam accounts for more than 10% of the total energy consumption of enterprises.Most of the existing operation dispatching methods rely on manual experience,and the management and control are extensive and the dispersion is huge.Therefore,it is of great significance to study the modeling and optimal scheduling of steam systems for reducing production costs and ensuring operational safety.Aiming at the operation characteristics of steam equipment layer and system layer,this paper proposes a data-driven dynamic modeling method for the gas-heat coupled equipment and a multi-scale optimal scheduling method for the steam system,respectively.At the equipment level,considering the coupling relationship between the input gas and output steam of the gas boiler,a factor modeling method based on the input delay LSTM-GRU is proposed.In view of the time-delay problem caused by the different response time scales of the two kinds of energy media,the dynamic time warping and causal probability are combined to calculate the delay parameters of the influencing factors of boiler steam,and samples are constructed respectively based on this,and a dynamic prediction model of steam generation flow based on Long Short Term Memory(LSTM)and Gated Recurrent Unit(GRU)is established.At the system level,in view of the synergistic interaction between day-ahead optimization and intra-day dynamic scheduling of steam system,a multi-time-scale dynamic optimization scheduling method is proposed.In the day-ahead optimization stage,the data samples are divided into granularities based on the production plan,an LSTM-based model is established to predict the steam production-consumption flow and pipe network pressure trends,and the optimization model is established with the goal of minimizing the economic cost of dispatching in 24 hours,and optimal pressure reference trajectories in each information particle interval are solved;in the intraday scheduling stage,a short-term pressure prediction method based on input delay LSTM is proposed,and a dynamic scheduling framework is designed to perform rolling optimization of the pressure in each information particle interval,realize the tracking of the reference trajectory of the day-ahead,so as to give reasonable scheduling suggestions.In order to verify the validity of the methods proposed in this paper,the actual operation data of a large domestic steel enterprises are used for experiments,which shows that the dynamic modeling method for the gas-heat coupled equipment proposed in this paper has higher accuracy than the commonly used back-propagation neural network modeling method,and the multi-time-scale scheduling method based on the rolling optimization has higher practicability and reliability than manual scheduling and event-driven scheduling methods,and can provide effective guidance for the decision-making of schedulers.