Single-inductor multi-port (SIMP) converters have great application potential due to their “more silicon less magnetic” feature. However, there are so many switching modes of SIMP converters that the modulation strategy design is complicated. At present, the design of the modulation strategy is mainly to manually select the switching mode sequence and carry out modal analysis, which requires professional knowledge of power electronics. This research proposes a reinforcement learning (RL) based modulation strategy design technique for SIMP converters. The proposed method uses a neural network (NN) to produce modulation strategies, takes known conditions such as port voltages and converter structure as the input of the NN and leverages a set of basic principles to provide rewards for training the NN, avoiding complex artificial design. With RL, the NN sums up experience during trial-and-error without human intervention, and finally generates the optimal modulation strategy under different operating conditions. Based on this method, modulation strategies are designed for a SIMP converter and the effectiveness of the method is verified by experiments.
- Article type
- Year
Open Access
Issue
A solid-state fault current limiter (SSFCL) is the key protective equipment in a direct current distribution network. In order to meet the high voltage requirements and reduce costs, implementing a SSFCL based on series-connected insulated gate bipolar transistors (IGBTs) is a promising approach. However, voltage unbalancing of IGBTs would be introduced if the gate-loops of the IGBTs are non-identical. In this paper, a plug-in gate-loop compensator with discrete gate voltage feedback and pulsewidth current compensation is proposed. The main merits are: 1) with the plug-in structure, the extra current sources only provide small power to fine-tune the gate-loop without affecting the functions provided by the commercial IGBT gate driver; 2) the gate-emitter voltages of IGBTs are compared with the preset thresholds to obtain control criterion, and the pulsewidths of the current sources are controlled for gate-loop compensation, thus both analog-digital and digital-analog converters are avoided; 3) the control law is easy to implement in FPGA, and is robust to voltage variation of power-loops. With the proposed compensator, the voltage unbalancing is alleviated immediately at the present switching cycle, and further eliminated cycle-by-cycle during the current limitation process. Experimental results verify the feasibility of the proposed compensator.
京公网安备11010802044758号