Design and Implementation of an Adaptive Synchronous Reference Frame Phase-Locked Loop Based on a Recurrent Neural Network

An adaptive self-tuned synchronous reference frame phase-locked loop (PLL) for a grid-following voltage source converter (VSC) is presented. The adaptation is conducted by a recurrent neural network (RNN) that predicts the optimized control parameter online based solely on already existent timeseries data. The resulting configuration adjusts its behaviour according to changing environmental conditions. An offline platform employs a genetic algorithm (GA) to extract the optimized control parameters for various operation points, which are then used to train the RNN model. The design goal for the optimization combines the damping of the PLL output with efficient frequency tracking. The conducted simulative case study demonstrates that the proposed implementation outperforms the conventional PLL in terms of robustness to varying conditions, such as alternating grid strength and load steps at weak grid scenarios. The approach is further validated through experimental measurements on a prototype laboratory setup demonstrating the real-time capability of the concept.