Data generation for digital simulators of metallurgical process operators

Abstract

The article deals with the formation of model implementations of time series of data (based on in-situ data) of controlled and uncontrolled impacts in simulator-training and digital modeling systems. Such simulators are becoming increasingly widespread due to the development of information and computer technologies, automated research systems, training systems, digital modeling technologies (APM modeling), as well as digital counterparts and advanced control systems. The formed implementations of impacts can characterize situations of normal process flow, emergency and pre-emergency states, or specific representative situations for training operators and technological personnel, software testing, research and tuning of algorithms and search for optimal control actions. Using examples from the metallurgical industry, the possibility of forming several interrelated impacts based on models of nonlinear dynamics and multivariate dynamic databases is shown. The Lorentz system describing the thermal convection of a fluid medium is considered as a model of the impacts formation. The model parameters for the low- and high-frequency components are determined separately, by processing in-situ data. Next, a training sample is formed using normalization and relay-exponential smoothing operations. The implementations of the actions are formed taking into account the mutual correlation of data based on models of chemical dynamics and are adjusted to the specified properties on a limited sample of a given volume with the required accuracy using a generator in the form of a closed dynamic system. The generator in form of a closed dynamic system is built on the basis of a multidimensional generating autoregressive model with adjustable coefficients. An example of the formation of data series on technological parameters of a blast furnace (the degree of wear of the furnace lining, temperature sensor readings and heat flux density) is shown.