부산대학교 도서관

This paper compares the effectiveness of different machine learning control algorithms for smart DC-AC power inverters that can dynamically adapt to the electric grid environment. Different algorithms are evaluated for varying input conditions, and performance metrics are analyzed. The control algorithms are evaluated with different techniques, including linear regression, polynomial regression, gradient regression, lasso regression, ridge regression, decision tree regression, elastic regression, random forests regression, and artificial neural network. The results of this study indicate that for DC-AC conversion, most basic regression tools that rely on polynomial equations or some variation of linear regression do not accurately predict the data; logical systems involving multiple situational regressions, or run multiple layers of calculation like random tree regression, and neural networks, are able to predict the data more accurately but at the cost of a time multiplier of up to 10-100 times longer. However, there is some potential for combining faster polynomial and linear algorithms to achieve a faster more reliable result when promptness is needed.