Prediction of dissolved oxygen content during RH refining process based on machine learning method

Factors affecting dissolved oxygen content were numerous. Conventional mechanistic or empirical models tended to have prediction errors that hindered the achievement of high-precision control. It was of great significance for establishing an accurate prediction model to reduce production costs. Accurate prediction of dissolved oxygen contents can be achieved by establishing appropriate machine learning models. The machine learning method including neural network technology, genetic algorithm and mind evolutionary algorithm was applied to predict the dissolved oxygen content and obtain the optimal slag addition amount during the refining process. Four types of prediction models were established to achieve the optimal model. The root means square error and mean absolute error were utilised to evaluate the accuracy. Impact factors were considered to determine the influence degree of each operating factor on the oxygen contents. The optimal slag addition amount with various T.AI contents was also calculated.