Scholarlink

<p>In this paper, a nonlinear model of an induction machine (IM) based on the black-box approach is presented, with the aim of accurate prediction of key output quantities of speed and current under variable supply and load conditions. The proposed model uses a feedforward artificial neural network (FFANN), whose weight coefficients are optimized by the Levenberg–Marquardt algorithm, based on data obtained by applying step changes of the voltage-to-frequency ratio (<i>U/f</i>) and load torque. The model does not require prior knowledge of the parameters of the equivalent circuit of the machine and is capable of generalizing complex nonlinear behavior for different machines and operating regimes. Validation was carried out experimentally on a real induction machine in laboratory conditions, as well as in a real-time simulation environment via the Typhoon HIL platform (HIL 101). The achieved errors during validation on the experimental machine in speed prediction are low (mean absolute percentage error—MAPE up to 0.34%), while the error in current prediction is negligible (MAPE ≈ 0.01%). The model was compared with other models, including recurrent neural networks with long short-term memory (LSTM) and nonlinear autoregressive models with exogenous input (NARX), where the FFANN showed superior performance in terms of accuracy and stability. Based on the obtained results, it can be concluded that the proposed model shows high accuracy and strong agreement with experimental measurements, significantly outperforming the performance of other considered nonlinear models.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Nonlinear black-box modeling of an induction machine under variable frequency and load conditions

  • 期刊论文

  • 发表日期：

    2026年2月17日

查看全文
  • Ilija Knežević,
  • Martin Ćalasan,
  • Tatijana Dlabač,
  • Mihailo Micev,
  • Filip Filipović,
  • Nebojša Mitrović

摘要

In this paper, a nonlinear model of an induction machine (IM) based on the black-box approach is presented, with the aim of accurate prediction of key output quantities of speed and current under variable supply and load conditions. The proposed model uses a feedforward artificial neural network (FFANN), whose weight coefficients are optimized by the Levenberg–Marquardt algorithm, based on data obtained by applying step changes of the voltage-to-frequency ratio (U/f) and load torque. The model does not require prior knowledge of the parameters of the equivalent circuit of the machine and is capable of generalizing complex nonlinear behavior for different machines and operating regimes. Validation was carried out experimentally on a real induction machine in laboratory conditions, as well as in a real-time simulation environment via the Typhoon HIL platform (HIL 101). The achieved errors during validation on the experimental machine in speed prediction are low (mean absolute percentage error—MAPE up to 0.34%), while the error in current prediction is negligible (MAPE ≈ 0.01%). The model was compared with other models, including recurrent neural networks with long short-term memory (LSTM) and nonlinear autoregressive models with exogenous input (NARX), where the FFANN showed superior performance in terms of accuracy and stability. Based on the obtained results, it can be concluded that the proposed model shows high accuracy and strong agreement with experimental measurements, significantly outperforming the performance of other considered nonlinear models.