The triple active bridge (TAB) converter is considered for connecting the different voltage power networks. The circuit circulating current will increase following the voltage difference increase. We propose machine learning (ML) aided modulation to improve the efficiency of the transmission power by suppressing circulating currents in the TAB converter. In the proposed modulation, it is necessary to identify the optimal operating point by varying five control variables. The five variables are the three duty ratios of the inverter voltage and the two phase shifts between ports. The five variables are determined by the port voltages and the input/output powers at each port. However, it is difficult to determine the five control variables mathematically to identify the optimal operating point. Therefore, in the proposed method, an ML model is used to determine the five variables. The model aims to find the optimal operating point based on the condition suppressing circulating currents. Experiments were conducted under the input voltage of 400 V and two output voltages of 250 V while varying the power sharing ratio and load factor ratio. The experimental results show an improvement in the efficiency of about 20.0 % under the light load condition of P2 = P3 = 0.2 kW compared with the general single phase shift modulation, as well as maximum efficiency of about 96.5 % under the full load condition of P2 = P3 = 2.0 kW.
Paper
Full text
Machine Learning Aided Optimized Modulation in Triple Active Bridge Converter
Semantic Scholar · Engineering · 2022
Abstract
The triple active bridge (TAB) converter is considered for connecting the different voltage power networks. The circuit circulating current will increase following the voltage difference increase. We propose machine learning (ML) aided modulation to improve the efficiency of the transmission power by suppressing circulating currents in the TAB converter. In the proposed modulation, it is necessary to identify the optimal operating point by varying five control variables. The five variables are the three duty ratios of the inverter voltage and the two phase shifts between ports. The five variables are determined by the port voltages and the input/output powers at each port. However, it is difficult to determine the five control variables mathematically to identify the optimal operating point. Therefore, in the proposed method, an ML model is used to determine the five variables. The model aims to find the optimal operating point based on the condition suppressing circulating currents. Experiments were conducted under the input voltage of 400 V and two output voltages of 250 V while varying the power sharing ratio and load factor ratio. The experimental results show an improvement in the efficiency of about 20.0 % under the light load condition of P2 = P3 = 0.2 kW compared with the general single phase shift modulation, as well as maximum efficiency of about 96.5 % under the full load condition of P2 = P3 = 2.0 kW.