Improved transient response for wide input range of DC-DC converter with neural network based digital controller

One of the important role for dc-dc converters is to regulate output voltage against the transient variation of input voltage. Recently, renewable energy becomes popular and renewable power generators are connected to the dc power grid directly, dc bus voltage is affected from them. Therefore, the improvement of the transient response against the variation of the dc bus voltage is required to control methods for dc-dc converters. The dc bus variation as the input voltage for dc-dc converters tends to be complicated, a flexible control method is required to improve the performance in the transient state. For such purpose, the digital control technique has potential to realize flexible control methods. In this study, to improve the transient response of a dc-dc converter against such wide input voltage variations, we present a novel neural network based digital control method. The neural network is adopted in coordination with a popular PID control to realize a flexible control by learning the non-linear input-output relation under various input voltage variations. It is well known that the neural network computation causes a computational burden for a widely used digital control unit such as digital signal processor. To address this problem, two different control periods, a slower one for the neural network computational process and a faster one for the conventional PID control process, are used to avoid computational burden in the digital controller in the presented method. Since the neural network works as a predictor, its prediction can be used for the combination with the PID control in the same timing of the faster control period. The evaluation result using a buck type dc-dc converter as a test study shows that the presented method improves the transient response effectively compared with the conventional PID control.