Optimized Switching Strategy for Multilevel Inverter Using Machine Learning for Minimum Voltage Distortion

Abstract

Multilevel inverters (MLIs) must be accurately modelled and predicted to enhance output quality, minimize total  harmonic distortion (THD). The three DC voltage sources are set up in the following ratio like 1:2:4 to allow the inverter to  produce different voltage levels using various switching combinations. By controlling the above seven switches and three  diodes properly, the above sources will be combined in such a way that a stepped output waveform having better resolution  and lower harmonics will be achieved. This configuration improves the quality and efficiency of the output for the supply of  an R-L load, and ensure efficient operation in applications of renewable energy. In this study, the ability of three advanced  machine learning algorithms namely Artificial Neural Network (ANN), Random Forest (RF) and Extreme Gradient  Boosting (XGBOOST) to predict MLI output under different operating conditions is investigated. Such kinds of data-driven  modelling models deal with the characteristics of inverter systems, namely the non-linear and dynamic system behaviour.   A comparative analysis, based on the accuracy of prediction and the performance of the generalization of the results, shows  that while ANN and Random Forest algorithms show an acceptable level of accuracy, the XGBOOST algorithm shows to be  better at performing than both models. The following ANN model R²= 0.9711 was obtained, which is 97 % of the variation  of the output, during the training of the neural network. Random Forest model showed better generalization with the  R²=0.9977 (training) and 0.9874 (testing), which is a good predictive capability. However, XGBOOST showed better  performance (R²=0.9999 during training, R²=0.9830 during testing indicating close to perfect learning of the training data  and excellent generalization on unseen data). Overall, the results present a good confirmation of the suitability of  XGBOOST as the most accurate and robust model with computational efficiency to predict the output voltage and harmonic  behaviour at the inverter power operation. It's reliability and speed make it a great tool for intelligent MLI control, real-time  fault diagnosis and integration as a part of renewable energy systems.