Implementation of Neural Network and Bias Correction Controls for Fast Transient Response of DC–DC Converter

This study presents a neural network-based control method of dc–dc converter to realize fast transient response in coordination with the PID control. The neural network control effectively improves the transient response by modifying the reference value in the PID control dynamically using its predictions in the transient state. However, it causes an overcompensation phenomenon and the convergence property becomes worse. To suppress the overcompensation and obtain the faster convergence, neural network timing control and bias correction are also adopted simultaneously. In the presented method, to realize edge-level prediction computation processing of the neural network control, data acquisition and repetitive training of neural networks are proceeded offline and the prediction by repetitive trained neural networks is implemented online in a commercially available computation unit. Experimental results confirm that the presented method obtains the significantly faster transient response compared with the conventional PID control.