Planning and scheduling of steel plates production. Part I: Estimation of production times via hybrid Bayesian networks for large domain of discrete variables

Knowledge of the production loads and production times is an essential ingredient for making successful production plans and schedules. In steel production, the production loads and the production times are impacted by many uncertainties, which necessitates their prediction via stochastic models. In order to avoid having separate prediction models for planning and for scheduling, it is helpful to develop a single prediction model that allows us to predict both production loads and production times. In this work, Bayesian network models are employed to predict the probability distributions of these variables. First, network structure is identified by maximizing the Bayesian scores that include the likelihood and model complexity. In order to handle large domain of discrete variables, a novel decision-tree structured conditional probability table based Bayesian inference algorithm is developed. We present results for real-world steel production data and show that the proposed models can accurately predict the probability distributions.