Digital Twin Analytic Predictive Applications in Cyber-Physical Systems

The article shows the relevance of the use of predictive models of digital counterparts for the formation and analysis of time trends obtained from the sensors of an automated control system. The requirements for the predictive model are shown; machine learning algorithms, regressions for time series forecasting are described; analysis and comparison of algorithms based on RMSE, MAE, R2 error readings are presented. Also, the article shows methods of automatic determination of emissions and novelty in time series and methods of detection of dependencies between parameters are brought. The authors give an example of integration of the predictive model into the infrastructure of a digital double, describe the life cycle and full functionality of such a system. In conclusion, the prospects of using the predictive model in systems where it is difficult to read the necessary parameters with low frequency are shown.