| Abstrakt: |
The article analyzes, generalizes, and develops the experience in solving predictive analytics problems that significantly affect traffic safety. New quality metrics for evaluating the operation of the model, as well as using hidden signs of operation of the traction motor for evaluating its condition, and a combination of various machine-learning methods can improve the performance quality of the models. The model operation stages are considered: a method of forming attributes and a set of initial data are shown, a quality metric is selected to evaluate the operation of the algorithm and observation intervals, various methods of constructing an algorithm for evaluating the condition of traction motors are proposed, and the results of their operation are compared with the real data. Possibilities of using the results of running the auto-encoder and expanding the scope of previously introduced metrics within the framework of newly introduced ones are shown, as well as the importance of using not only the results of electrical measurements, but also diagnostic data, as input data. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink