Research on Scheduling Method for Uncertainty of Hit Rate of Molten Steel Based on Q Learning

Abstract

Aiming at the problem of uncertain scheduling of molten steel hit rate in the steel refining process, taking into account the multi-stage, multi-equipment, and multi-constrained production process conditions of refining production and the process of refining process due to the uncertainty of molten steel hit rate during the refining process, in order to obtain a scientific and feasible approximate optimal scheduling plan in a short period of time, the system state and system state transfer rules of the steel production process are defined, and the random evolution scheduling optimization system model of steel production refining based on the discrete-time Markov chain is established. At the same time, in the refining process scheduling optimization problem, the complexity of the solution will increase exponentially with the increase of the number of reprocessing processes, and a stochastic dynamic programming algorithm based on heuristic simulation strategy and improved Q learning is designed to solve the problem. Aiming at the uncertain scheduling problem of molten steel hit rate under different process production paths, simulation experiments using actual production data of a large domestic steel mill verify the effectiveness of the proposed model and algorithm.