| As one of the main industries of national economy, iron and steel production is of great importance for which the energy conservation and diffusion control is directly related to the international competitiveness of the enterprises and the goal of green manufacturing. Regarding the whole production flow, steel-making process acts as the primary phase. And the prediction and scheduling work on corresponding gaseous energy system plays a vital role for the development of the whole steel industry. With the advancement of related research and study, new demands are emerging which typically on acquiring broader prediction horizon, multiple dimension of the output and higher efficiency of scheduling. Take this into account along with practical configurations of the gaseous energy system in steel-making process, i.e., Linz-Donawitz Converter Gas (LDG) system and oxygen/nitrogen system, this study proposed some prediction and scheduling methods as follows.Aiming at predicting the long period of future trend on the generation and consumption amount of LDG, oxygen and nitrogen, a Granular-computing (GrC) based long-term prediction model is consequently proposed. Considering the periodic features in production, the approach divided the data into granules (in unequal length) so that the analysis horizon is extended from pointwise to segment-formed. For accommodating with the following clustering algorithm, time-warping path is then utilized here for normalizing the granules as in equal length. And by means of fuzzy c-means clustering, fuzzy inference and centroid defuzzification techniques, the long-term prediction results are finally obtained.In view of estimation on the multiplestorage amounts of LDG system, a multi-output least square support vector regressor (MLSSVR) and a GrC based hybrid collaborative fuzzy clustering (HCFC) model are established as short and long-term prediction methods respectively for the multiple gas tank levels.The MLSSVR involves not only the single fitting error of each tank level but also the combined one so as to improve the prediction accuracy. The hybrid structure of HCFC considers the features regarding to a gas tank, of which the horizontal part elaborates the mutual influences among different time spaces of a tank level, while the vertical one describes them among the influence factors (denoting the gas generation units or the users). Furthermore, GrC analysis pattern is deployed here so that the prediction length is prolonged to more than 480 points (i.e.,8 hours) which realized long-term prediction.For the balance and optimization work on oxygen/nitrogen system, a Mixed Integer Linear Programming (MILP) based scheduling model is proposed. On the basis of long-term prediction results of the requiredamounts on oxygen and nitrogen, the model takes gaseous energy diffusion rate as the objective, while the actual capacity of the devices along with the practical energy conversion procedure is formulated as the constraints. Besides, the overall scheduling flow value and equipment amounts are restricted for making the computed plan convenient and applicable which avoids the problems of over scheduling, excessive amounts of scheduled equipment, etc.Utilizing real data of gaseous energy system of a steel plant in China, the experiments on the above methods are thoroughly conducted which clearly stated the superior performance of single/multiple output short/long term prediction on the generation/consumption and storage amounts. It can be also proved that the scheduling model successfully reduces the diffusion rate and has higher efficiency comparing with the manual approach.The applied results of the developed software system are also given for demonstrating the practical effectiveness of the proposed models. Such work can be regarded as strong support which is beneficial for achieving efficient steel production and energy utilization. |
