Computer system and method providing operating instructions for thermal control of a blast furnace

A computer-implemented method for training a reinforcement learning model to provide operating instructions for thermal control of a blast furnace, the method comprising:

• processing, by a domain adaptation machine learning model trained by transfer learning, historical operating data obtained as multivariate time series and reflecting thermal states of respective blast furnaces of multiple domains, to generate a first domain invariant dataset representative of the thermal state of any of the blast furnaces irrespective of the domain;
• generating artificial operating data as multivariate time series reflecting a thermal state of a generic blast furnace for a particular thermal control action by using a transient model of the generic blast furnace process, wherein the transient model reflects respective physical, chemical, thermal, and flow conditions of the generic blast furnace, and provides solutions for upward gas flow and downward moving of solid layers as structured in the generic blast furnace while exchanging heat, mass, and momentum transfer;
• processing the artificial operating data by a generative deep learning network trained on multivariate time series of the historical operating data, to generate a second domain invariant dataset by transferring the features learned from the historical operating data to the artificial operating data;
• the reinforcement learning model determining a reward for the particular thermal control action in view of a given objective function by processing the combined first and second domain invariant datasets;
• and
• dependent on the reward, re-generating the second domain invariant data set based on modified parameters, wherein a genetic search and/or a Bayesian optimization algorithm is guiding the search for the modified parameters for a further thermal control action based on the current environment of the reinforcement learning model and the thermal control action output of the current learning step, and repeating the determining step to learn optimized operating instructions for optimized thermal control actions to be applied for respective operating states of one or more blast furnaces.