Machine learning in the development of regulators: the specifics of data preparation and the choice of the ML model algorithm

The article considers the stages of regulators development based on machine learning. Machine learning, like a fuzzy logic, is used in PID regulators in two ways: it is able to build the regulator or a unit for adjusting its coefficients. A feature of machine algorithms is the “learning” ability which allows you to transfer the experience of an expert to the system. The presence of a learning procedure distinguishes the configuration of regulators based on machine algorithms from the classical approach. The publication analyzed the possibility of using regression machine algorithms (models) of the Scikit-learn library (Python) to solve control problems using the example of a level regulator. The search for an effective data processing model led to the construction of ensemble algorithms based on stacking. The high degree of compliance between predicted and test data confirms the accuracy and stability of the results. The results presented in the article can be used in a simulating or in developing a controller based on a machine algorithm.