Adaptive Frequency Control of DC-DC-Converters for Maximum Efficiency Using Artificial Neural Network

Two-Quadrant-Converters (2QC) are widely used for many applications. Such a 2QC is commonly operated at a constant switching frequency or a switching frequency which leads to a constant operation mode. However, the selected switching frequency does not lead to minimal total losses for the entire operating range. The loss-optimal switching frequency strongly depends on the operation point so that the operation with constant switching frequency would lead to higher losses than necessary. In this paper, the switching frequency is adapted depending on the operating point in order to achieve minimal total losses. The loss-optimal switching frequency depends on numerous parameters, hence storing the data in a look-up-table could suffer from huge memory demand. To overcome this problem, an artificial neural network is trained. A lab-prototype was built in order to verify the benefits achieved by the proposed strategy. Finally, the artificial neural network is embedded in a DSP-based highdynamic control. Experimental results show a reduction of total losses up to 25%.