Characteristics study of neural network aided digital control for DC-DC converter

This paper studies the digital controlled dc-dc converter which employs the neural network in coordination with PID control. Especially, we focus on the problem of overcompensation to the transient response of the output voltage in the DC-DC converter by the neural network control. Our previous studies suggested that the neural network control is effective to improve the transient response of the output voltage. Therefore, it can contribute to the low power consumption of power supplies for information and communication systems. However, the neural network aided control has a problem that it tends to be overcompensation, i.e., the compensation effect of the neural network becomes too much and badly affects to control the plant. The overcompensation causes from a delay time in sensing and calculation to determine the timing and duration of the neural network control itself. To address the problem of overcompensation, our previous studies investigated methods to devise the timing and duration of neural network control. However, since they need some pre-determined parameters which are optimized to the specific target converter to avoid the effect of delay, the implementation of it tends to be complicated one in practical use. In this study, we adopt the neural network which predicts parameters to improve the transient response and control timing and duration of itself. Therefore, it does not need any pre-determined parameters for timing and duration control. From the simulation result for evaluation, it is confirmed that the presented method has a superior transient characteristics compared with conventional method.