Application of artificial intelligence in dual-active-bridge (DAB) converters

In recent years, with the development of power electronics in power systems, DC microgrid systems based on new energy sources have become important for research [1]. For the collection, conversion, and transmission of new energy sources, the Dual-Active-Bridge (DAB) DC-DC converter has become a key energy conversion device. The DAB DC-DC converter was first proposed in the early 1990s by Rik De Doncker and Deepakraj M. Divan et al. [2,3]. It contains an isolation transformer T r, an external series inductor L r, and two bridge converter units (Full bridge 1 and Full bridge 2) as shown in Figure 4.1. The power delivery of DAB converters can be both bidirectional and unidirectional; thus, they are widely used in renewable energy power generation systems, Solid State Transformer (SST), Energy Storage Systems (ESS), Electric Vehicle (EV), and aerospace because their simple structure, wide range of soft switching, reliable performance, and high power density are achievable [1,4,5].However, the mathematical model of the DAB DC-DC converter is complicated due to the large number of active and passive components in the converter. In addition, because the relevant parameters and external circuit environment in the DAB DC-DC converter may change with the change in operating conditions, it is difficult to model it accurately using the traditional mathematical modeling approach. Since the DAB DC-DC converter model has many adjustable control variables, it aggravates the computational effort and complexity of the efficiency optimization solution for this type of converter. The existing numerical optimization methods, iterative algorithms, and heuristic algorithms have the drawbacks of complex models, time-consuming computation, and difficulty in finding the optimum, which make it difficult to ensure fast and comprehensive optimization for each optimization objective [1,6,7]. Therefore, how to improve the efficiency and performance of DAB DC-DC converters has become important in the field of power electronics.Due to it being a high-dimensional system for optimization with complex and changing models, the new generation of artificial intelligence techniques has shown their advantages in many systems with large amounts of data, complex modeling, and uncertainty in optimization decisions, and they are well-suited to solve the optimization problems of high-dimensional complex systems [8,9]. Currently, these methods have demonstrated superior performance in finding and making decisions in other areas of electrical engineering (e.g., power systems, and power markets). Currently, reinforcement learning methods are rarely used in the field of power electronics optimization and are therefore well worth exploring [10,11]. For power electronic converters with complex parameters like DAB DC-DC converters, it is of theoretical and practical value to deeply integrate artificial intelligence algorithms and study the key bottlenecks of reinforcement learning to be used in DAB DC-DC converter efficiency optimization.Based on the above background, this chapter will delve into the problem of optimal modulation of reinforcement learning algorithms in DAB DC-DC converters to improve the performance and transmission efficiency associated with DAB DC-DC converters.