Efficiency is one of the key elements to characterize the performance of inverter-fed permanent magnet synchronous machines (PMSMs). However, it is challenging to establish an efficiency model that accurately characterizes the input-dependent efficiency uncertainty. Therefore, this article proposes to train a novel heteroscedastic Gaussian process regression (HGPR) model for quantifying the input-dependent uncertainty in inverter-fed PMSM efficiency modeling. Based on experimental data from a 70-kW inverter-fed PMSM, statistical characteristic analysis shows that the efficiency data at each operating point is approximately normally distributed. Then, HGPR probabilistic inference is derived by applying input-dependent heteroscedastic noise and variational approximation. Estimation results show that compared with the Gaussian process regression (GPR) model, the HGPR model gains high accuracy in terms of efficiency mean and significantly improves the coverage-width-based criterion (CWC) by 21.78% and continuous ranked probability score (CRPS) by 10.72% in terms of probabilistic distribution. Further analysis indicates that the improvement in HGPR efficiency estimation is attributed to the input-dependent noise term. The application to a driving cycle of a pure-electric vehicle powertrain validates the superiority in the estimation of efficiency and electricity consumption.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    A Bayesian Probabilistic Approach for Quantifying Input-Dependent Uncertainty in Inverter-Fed PMSM Efficiency Modeling


    Beteiligte:
    Guo, Fengxiang (Autor:in) / Chen, Li (Autor:in)

    Erschienen in:

    Erscheinungsdatum :

    01.12.2024


    Format / Umfang :

    3172874 byte




    Medientyp :

    Aufsatz (Zeitschrift)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch